Huawei SAP HANA Solution

Pre-sales Training
After this course, you should be able to answer the following questions:

1 Huawei SAP HANA solution supports a maximum of ( )-socket.

2 The maximum ( ) TB of RAM is supported by a single Intel Cascade Lake or Coopler Lake CPU in
OLTP scenario, for example SAP S/4HANA system.

3 The maximum ( ) TB of RAM is supported by a single Intel Cascade Lake or Coopler Lake CPU in
OLAP scenario, for example SAP BW/4HANA system.

4 Huawei Kunlun 9008 V5 HANA appliance supports maximum ( ) TB RAM. And support maximum ( )
TB RAM is using Intel Optane Persistant memory DIMM.
Contents

• SAP and SAP HANA Introduction

• Huawei SAP HANA Solution
• Huawei SAP HANA Solution Highlight
• Huawei SAP HANA Configuration and Quotation Guide
• Huawei SAP HANA Resource and Project Case

3 Huawei Confidential
SAP - The Largest Provider of Enterprise MGT Solutions

SAP is "Systems Applications and Products in Data

Processing." It has two meanings:
First, SAP is the company name. It is the world's largest
enterprise management and collaborative e-commerce
solution provider based in Waldorf, Germany, and the
third largest independent software provider in the world.
Second, SAP is the name of its software. It is the pioneer
of ERP solutions and the world's number one ERP
software. It can provide comprehensive solutions for
enterprises of all sizes and sizes.

420,000 79,960 $29.18b 87%

Customers in 190 Employees in 130+ Annual revenue(IFRS) Of Forbes Global 2000
countries countries in FY2018 are SAP Customer

47years 100+ 12,100 1.1亿

Of history and innovation Innovation and SAP Partner companies Subscribers in SAP Cloud
Development centers globally User base
4 Huawei Confidential
SAP Main Products

SAP APPs
Hybris BO … Concur
Ariba
E-commerce Report presentation Travel management
platform Subscription

BW, Customer relationship mgmt

Application BW/4HANA CRM C/4HANA
Tier PLM Customer Relationship
Business warehouse SRM
ERP, Mgmt
Product S/4HANA Supplier
lifecycle mgmt
EWM relationship mgmt Successfactor
SCM
Extended warehouse Human resources
Supply chain mgmt
mgmt
Enterprise Business Suite

Database
Tier SAP HANA Database
5 Huawei Confidential
SAP Products Evolution
• SAP helps enterprises of different industries and scales to optimize their operations. 77% of the global transaction
revenue is related to the SAP system.
• SAP HANA is an in-memory data platform that was officially launched in 2011. It is one of the SAP products with the
fastest user growth.
• For SAP Business Suite that is not deployed on the SAP HANA platform, SAP plans to terminate the
maintenance in 2027.

BW on
BW on HANA
R/3
R/3 HANA Suite on
Client/Server Suite on HANA
SAP R/2 Client/Serve
non-
Application R/1
R/1 Mainframe HANA
Layer Mainframe
Mainframe
1972 1979 1992 2012 2013 2015 2016
Database Non- Non- SAP SAP SAP
Non-HANA Non-HANA Non-HANA
Layer HANA HANA HANA HANA HANA

6 Huawei Confidential
SAP Enterprise Software Introduction
Category Type HANA Description
SAP's latest generation only supports HANA's business suite, which can be understood as a new generation of SAP ECC, based on HANA for
S/4HANA Yes development and optimization;
Enterprise Resource Planning include modules such as financial accounting, financial management, accounting management, enterprise control,
investment management, production planning, material management, factory maintenance, quality management, project management, sales and
ERP Yes distribution management, human resources management, and open information warehouse. ECCs can be used to build a world -class management solution
to support enterprise operation.
Supplier Relationship Management (SRM) provides support for activities such as expenditure analysis, goods source development , contract management,
SRM Yes procurement, invoice issuing, and supplier management. SRM system can be seamlessly integrated with SAP ECC system to unify d ata.
Master Data Management (MDM), an important part of the SAP NetWeaver platform, provides powerful data cleansing and centralized data management.
MDM No This helps enterprises implement unified and efficient management of customer or product master data. MDM can perform data co nsolidation to reduce
data maintenance costs, ensure inter-system data consistency, and accelerate service processing.
Business
Process Integration (PI) is an integration technology and platform used by ERP for exchanging information between various com ponents in the ERP system
Suite PI/PO No in a synchronous or asynchronous manner.
CRM Yes Indicates the Customer Relationship Management (CRM) system.
(OLTP) HRM Yes Indicates the Human Resources Management (HRM) system.
EWM Yes Indicates the Extended Warehouse Management (EWM) system.
Web Dispatcher is a core component for load balancing in the SAP system. It is deployed in the demilitarized zone (DMZ) and p rovides only one ingress for
Web
No users. Web Dispatcher needs to connect to an SAP message server before configuration. Web Dispatcher ensures data access security. It can be used as
Dispatcher an access port for future SRM suppliers.
NW Gateway is a gateway in the SAP software system and is crucial for the CPI-C service. This gateway is based on TCP/IP. All requests for change
NW
No (RFCs) are sent though the SAP gateway. In the SAP system, each instance has one SAP gateway. NW Gateway is used for data transmission and
Gateway publishing in the SAP system.
Solution Manager is a software lifecycle management system that can be used to manage all SAP software platforms. It provides various management
Solution
No functions, including system monitoring, system upgrade, solution document management, and test management. SAP Solution Manag er is an important
Manager auxiliary system for installing SAP ERP software.

The systems highlighted with a red background are the most frequently used HANA systems, followed by the ones highlighted with yellow. Theoretically, all SAP
systems can be deployed on HANA databases. The "Yes" or "No" in the preceding table is just a recommended choice for customers based on considerations of the
best cost-benefits.
7 Huawei Confidential
SAP Enterprise Software Introduction (continue)
Category Type HANA Description

BW4HANA Yes BW/4HANA is an evolution of BW that is completely optimized and tailored to HANA.
Business Information Warehouse is an EDW product. SAP BW is another important enterprise information system launched after SAP ERP and is
believed to be the most functionally powerful data warehouse system. SAP BW supports SAP and non-SAP system consolidation, and offers a rich
Business range of data conversion tools, powerful staging mechanisms, easy-to-use front-end reports, and consolidated web reports. SAP BW uses various tools
BW/BPC/BI Yes to convert data for non-SAP systems, enabling analysts and decision makers to rapidly obtain highly-integrated information from different systems. SAP
Warehouse
BW also provides accurate comprehensive reports and business intelligence functions.
Business Planning and Consolidation (BPC) is an enterprise financial budget platform based on SAP BW.
(OLAP)
BusinessObjects (BO) is a business intelligence product. It includes BusinessObjects Enterprises, Crystal Reports, Webintelli gence, Crystal Dashboard
BO No Design, and SAP BusinessObjects Edge solutions, fully meeting enterprises' requirements for business intelligence data analyt ics. Generally, BO uses
the HANA database of BW.

The systems highlighted with a red background are the most frequently used HANA systems, followed by the ones highlighted with yellow. Theoretically, all SAP
systems can be deployed on HANA databases. The "Yes" or "No" in the preceding table is just a recommended choice for customers based on considerations of the
best cost-benefits.
8 Huawei Confidential
SAP Enterprise Software Layered Architecture

Usually using

Application
Virtualization
+

layer
S/4HANA BW/4HANA … SCM CRM
FusionServer OceanStor

Usually using
Physical Server
Database

SAP
layer

HANA

HANA Database SAP HANA Appliance/TDI

DR layer*
Backup&

HANA System
Backup
Replication
Backup DR storage
Huawei OceanStor

9 Huawei Confidential
SAP System Landscape

2-System Landscape Typical deployment architecture: DEV PRD

development and production

3-System Landscape Typical deployment architecture: DEV QAS PRD

development, testing and production

Increase the "pre-production" system to

4-System Landscape reduce overall system risk and increase DEV QAS PRE PRD
flexibility

Landscape Description High Availability

PRD Production System Y
DEV Development System N
QAS Quality& Assurance System N
Sand-Box Production mirror environment(every month or week) N
UAT/Pre-prod Environment is for business department users to do the test N
DR Disaster Recovery environment N
10 Huawei Confidential
SAP HANA Deployment

SAP HANA Service Modes SAP HANA Deployment Modes SAP HANA Architecture

▪Local ▪SAP HANA Solution 1: Solution 2: TDI (Tailored Solution 1: scale-up Solution 2: scale-
deployment Cloud Appliance Data center Integration) (single node) out (cluster)

End User ISP or

• Suite on HANA • BW /Analytics on HANA
Carrier • BW /Analytics on HANA

S/4HA
NA S4
CRM HANA

BW4H Hybris
AN Success
SRM Factor
Network
…
…

Network

SAP SAP Storage

HANA HANA
Shared
storage
Physical or Virtualized

11 Huawei Confidential
SAP HANA Deployment Architecture of Scale-up

SAP HANA Architecture SAP HANA Production System HA

Client App Server

DEV QAS PRD

Operation System HA Cluster (Virtual IP)
HANA agent

heartbeat
monitor monitor

Primary Secondary
(active) (active)
System (data pre-loaded)
Development Test Production HANA Replication HANA
system system system
Data Log Data Log
Disks Disks Disks Disks

Superior Second-level
Simple O&M
performance switchover
12 Huawei Confidential
SAP HANA Deployment Architecture of Scale-up with DR

Main DC Remote DR Center

System
replication

Synchronize
or
Primary Secondary Asynchronize Secondary
(active) (active) (active)
System (data pre-loaded) (data pre-loaded)
Replication
HANA HANA HANA

Synchronize
Data Log Data Log Data Log
Disks Disks Disks Disks Disks Disks

13 Huawei Confidential
SAP HANA Deployment Architecture of Scale-out

SAP HANA Architecture SAP HANA Production System HA

HANA HANA HANA

DEV QAS PRD master slave 1 … standby

HANA cluster
…

Storage Connection

/hana/shared

Development Test Production Log 0 Log 1 Log M

/hana/shared/<SID>
system system system Storage pool
…
Data 0 Data 1 Data M

Shared file system Block devices

14 Huawei Confidential
SAP HANA Deployment Architecture of Scale-out with DR

Business continuity
Local HA DR
Production Site (HA) DR Site
HANA HANA HANA HANA HANA
master slave 1 … standby System master M slave

replication
HANA cluster
…
… Disaster
recovery across
1000 km
Storage Connection
Storage Connection

/hana/shared /hana/shared …
Log 0 Log 1 Log M Log 0 Log N
/hana/shared/<SID>
Storage /hana/shared/<SID>
Storage pool pool
…
Data 0 Data 1 Data M Data 0 Data N

Shared file system Block devices Shared file system Block devices

15 Huawei Confidential
Contents
• SAP and SAP HANA Introduction
• Huawei SAP HANA Solution
• Huawei SAP HANA Solution Highlight
• Huawei SAP HANA Configuration and Quotation Guide
• Huawei SAP HANA Resource and Project Case

16 Huawei Confidential
Huawei SAP HANA Full-Series Solutions Meet Diversified
Requirements
Single-node solution Cluster solution TDI solution

KunLun 9008 V5 (8S)

FusionServer Pro & KunLun FusionServer Pro & KunLun

2488H V5/V6 (4S)

2288H V5 (2S)
OceanStor OceanStor/Dorado

➢ SAP Business Suite, such as SAP S/4HANA, ➢ SAP Business Warehouse on HANA and ➢ Applicable to customized deployment, such as
ERP on HANA, and CRM on HANA BW/4HANA, supporting massive scale-out deployment solutions with specific storage
➢ SAP Business Warehouse on HANA and ➢ S/4HANA, supporting scale-out requirements, and supports multiple services,
BW/4HANA, restricted by the memory size of a including HANA Cloud
single node ➢ E2E performance ensured by hardware vendors
➢ Non-SAP application systems, such as the third- and customers, not by SAP
party application running on HANA ➢ SAP Business Warehouse on HANA, Business
➢ SAP B1 on HANA Suite on HANA, BW/4HANA, and S/4HANA

17 Huawei Confidential
Sales Strategies for V5 Single-node Solutions

Single-node Solution Guidance Strategies:

➢ If the required capacity is lower than 1.5TB, recommend
KunLun 9008 V5 (2S/4S/6S/8S) 2288H V5 to customers who look for the best price and
do not require capacity expansion.
2488H V5/V6 (2S/4S) ➢ If the required capacity is 2.25 TB to 6 TB, 2488H
V5/V6 is recommended.
➢ KunLun 9008 V5 HANA (4S/6S/8S) is recommended, if
2288H V5 (2S)
Q3 the required capacity is more than 6 TB.
➢ Recommend customers to use the Optane persistent
memory solution, which can significantly reduce
192 GB … 1.5 TB 3 TB 4.5 TB 6 TB 9 TB 12 TB business cost *.
2S (56 cores) ➢ All-flash appliance solution is preferred.
4S (112 cores)
＊SAP HANA Optane Persistent Memory Solution
6S (168 cores) / 8S (224 cores)
http://3ms.huawei.com/documents/docinfo/1899371

18 Huawei Confidential
Sales Strategies for V5 Scale-out Solutions
Scale-out Solution

KunLun 9008 V5 for S/4HANA (8S)

60 TB*
KunLun 9008 V5 for BW/4HANA (8S) (4+1 nodes) Guidance Strategies:
➢ 2488H V5 HANA is recommended if the required
2488H (4S) V5/V6 96 TB capacity is 3 TB.
(16 nodes) ➢ KunLun 9008 V5 HANA (8S) is recommended if the
48 TB required capacity is 6 TB and larger per node.
(16 nodes)
➢ KunLun 9008 V5 HANA (8S) is recommended for
S/4HANA scale-out large capacity. It supports 8S/6T or
8S/12T S/4HANA 4+1 cluster
3 TB … 3 TB … 6 TB … 12 TB
4S (112 cores)
8S (224 cores)
8S (224 cores)

* S/4HANA Scale-out maximum 4+1 nodes, single node minimum size is 6 TB and maximum size is 12 TB

19 Huawei Confidential
SAP HANA TDI Solutions Support Larger Memory

TDI solution
Non-volatile memory to support larger memory of a single node

KunLun 9008 V5 (2S/4S/6S/8S)

2488H V5/V6 (2S/4S)

2288H V5 (2S)

1,920 GB ... 6 TB 9 TB 12 TB 15 TB 18 TB 22.5 TB 24 TB

2S (56 cores)
4S (112 cores)
6S (168 cores)
8S (224 cores)

20 Huawei Confidential
SAP HANA Persistent Memory: Faster Startup and Lower TCO

DIMM Capacity per

Speed
capacity processor
3 TB (excluding
128/256/512 GB 2,666 MT/s DRAM)

Intel® Optane™ Persistent Memory (PMem) Previous known as: Intel® Optane™ DC Persistent Memory (DCPMM)

HANA startup time slashed by 80%* More cost-effective

1600
1394
1400
1200
1000
800 80% 20%
600 Memory capacity/$

400 285
200
Larger memory capacity at the same price
0 • Providing the same memory capacity, the PMem solution slashes the CAPEX.
Unit: second 64G DRAM * 96 32G DRAM*48+128G PMem*48 • According to the performance test results in the Huawei HANA lab, the
* Huawei KunLun 9008 V5 supports 8 Intel® Xeon® Platinum 8276L processors and 5 x 3.2 TB SSD
performance of the PMem solution is slightly lower than that of the DRAM solution
RAID5, with 9 datasets of 3.1 TB data. with the same memory capacity. The performance decrease is less than 5%.

21 Huawei Confidential
SAP HANA Industry Distribution and Proportion

SAP HANA Industry Distribution and Usage Proportion

67% 93%
50%
87% 97%
86% 86% 78%
100% 100%
69% 100% 48% 37%

Fortune Top500 SAP Ratio

22 Huawei Confidential
Huawei and SAP Work for 8 Years to Boost Innovation and Build
Industry-Leading HANA Solutions

Huawei became SAP's Huawei and SAP jointly Strategic alliance partnership: Huawei and SAP signed a As a top-level sponsor, Huawei & SAP Co- SAP attended HUAWEI Huawei and SAP launched
first global technical launched Tecal RH5885 V2, Huawei released the certified MoU at CeBIT 2015 to Huawei attended the Innovation Center was CONNECT 2018 the FusionServer Pro V6
partner in China. the first SAP HANA appliance. SAP HANA FusionCube deepen collaboration in SAPPHIRE NOW 2016 global inaugurated in Zone G of as a diamond sponsor SAP HANA solution at
July 2012 Feb. 2013 solution at CeBIT 2014. Industry 4.0 and the IoT. flagship conference held by Huawei HQ, Shenzhen. and major partner. HUAWEI CONNECT 2020.
Mar. 2014 Mar. 2015 SAP in Orlando, USA. Aug. 2018 Oct. 2018 Sep. 2020
May 2016
MaxAttention for Hybris Billing
HANA SAP Hybris &

2012 2013 platform

adoption
2014 2015 2016 Ariba contract
signed 2017
Hybris Billing
LIVE @ CBG 2018 Rollout @ CBG &
Cloud 2019 2020

Continuously incubated innovative applications based on HIS-Onedata and HANA.

Ren Zhengfei, founder of Huawei Day was held at Walldorf, Huawei and SAP announced SAP conferred the d- Huawei won the SAP Huawei attended SAP China Huawei introduced the
Huawei, met with Jim Germany, the SAP headquarter. the official opening of the kom award on Huawei. HANA® Innovation Award Summit as a diamond persistent memory HANA
Snabe, Co-CEO of SAP. July 2014 Huawei & SAP Co-Innovation Dec. 2015 at SAPPHIRE NOW held sponsor to explore market solution for larger HANA
Sep. 2012 Center in Shenzhen. in Orlando. opportunities and innovation. memory capacity.
June 2015 May 2017 Sep. 2018 May 2019

SAP Max. Attention | SAP and Huawei Day | Huawei & SAP Co-Innovation Center
23 Huawei Confidential
Huawei Provides Professional Implementation Services
- SAP HANA Appliance
SAP HANA Appliance contains the OS and HANA database pre-installation by default*. But SAP HANA Scale-out Appliance
involves external storage links and network connection, must be installed at the customer site.
The parts provided by Huawei are marked in ORANGE,
including hardware installation, OS installation, HANA Integration and
database installation, and hardware maintenance. System maintenance Enterprise Development
application
1. Hardware installation.
2. OS installation: SUSE OS installation (pre-installation or
onsite installation is supported. Huawei resells the SUSE HANA HANA Installation&
license or customers purchase the SUSE OS); The HANA Databases High Availability
solution implementation service needs to be configured for Configuration
onsite installation.
Operating
3. HANA database installation: (pre-installation or onsite System
installation is supported, but Huawei does not resell HANA
licenses**); The HANA solution implementation service
needs to be configured for onsite installation.
Hardware Installation& Hardware
4. (Optional) HANA Solution Implementation Service: OS maintenance
installation on-site, HANA database installation onsite, and
HANA High Availability Configuration***
5. Huawei provides one-stop hardware maintenance services.

* The OS and HANA can be preinstalled only when 600 GB SAS 10k HDDs are used as the OS boot disk or the OS+HANA shared volume is merged.
** After the HANA database is installed, a temporary license period of 90 days is available
*** HANA HA configuration involves the network connection between the Active and Standby nodes and have to be implemented at the customer site.
24 Huawei Confidential
Huawei Provides Professional Implementation Services
- SAP HANA TDI
SAP HANA TDI includes OS preinstallation by default*

The parts provided by Huawei are marked in ORANGE

(Optional in GREEN), including hardware installation, OS
installation, HANA installation, and hardware maintenance. Integration and
System maintenance Enterprise Development
application
1. Hardware installation
2. OS installation: SUSE OS installation (pre-installation or
onsite installation. Huawei resells the SUSE license or
HANA
customers purchase the license by themselves.); The Databases
HANA Installation&
High Availability
HANA solution implementation service needs to be Configuration
configured for onsite installation.
3. HANA database installation: (Only onsite installation is Operating
System
supported. Huawei does not resell HANA licenses**) The
HANA solution implementation service needs to be
configured for onsite installation.
4. (Optional) HANA solution implementation service: OS Hardware Installation& Hardware
maintenance
onsite installation, HANA database onsite installation, and
HANA HA configuration***
5. Huawei provides one-stop hardware maintenance services.

* The OS and HANA can be preinstalled only when 600 GB SAS 10k HDDs are used as the OS boot disk or the OS+HANA shared volume is merged.
** After the HANA database is installed, a temporary license period of 90 days is available
*** HANA HA configuration involves the network connection between the Active and Standby nodes and have to be implemented at the customer site.
25 Huawei Confidential
Contents
• SAP and SAP HANA Introduction
• Huawei SAP HANA Solution
• Huawei SAP HANA Solution Highlight
• Huawei SAP HANA Configuration and Quotation Guide
• Huawei SAP HANA Resource and Project Case

26 Huawei Confidential
Challenges to SAP HANA Deployment

Innovative Huawei SAP HANA solution

Scalable
✓ Flexible scale-out to 2S/4S/6S/8S, supporting
appliance, cluster, and TDI solutions
Challenges for enterprises to ✓ Large capacity, scalable to 24 TB using PMem*
deploy SAP HANA databases Challenges
Innovative Swift
◼ Inefficient running solution ✓ Scale-up first, high performance, and easy
maintenance
◼ High costs ✓ No.1 in the SAP BWH Benchmark

◼ Insufficient memory capacity Stable

✓ Enterprise-class reliability
◼ On-demand expansion not
✓ Highest reliability in the industry (x86)
supported ✓ Diversified SAP HANA HA and DR solutions

◼ Complex O&M
Smart
✓ HANA RAS Enhancement Tool
✓ Automated Deployment Tool

* Pmem: Intel® Optane™ Persistent Memory

27 Huawei Confidential
 Scalable Swift Stable Smart

Huawei Provides the SAP HANA Solution with the Smallest

Expansion Granularity in the Industry

Smooth Single memory Higher

expansion specification reliability
KunLun 9008 V5 2488H V5/V6

2S/1.5 TB 4S/3 TB 6S/4.5 TB 8S/6 TB 2S/1.5 TB 4S/3 TB

2S/3 TB 4S/6 TB
2S/3 TB 4S/6 TB 6S/9 TB 8S/12 TB

Supporting smooth scale-up from 2S 4S 6S 8S, maximizing

return on investment (ROI).

28 Huawei Confidential
 Flexible High High
Smart
scalability performance reliability

Huawei SAP HANA Appliance Ranks Top 1 in SAP BWH Benchmark

SAP® BWH Benchmark for 4-socket servers
Query Executions per Hour
Top1
Top1 HANA
Huawei 2488H
HANA Benchmark
V6
7000 Benchmark
6766 5400
6800
6617 Huawei 2488H
5293
6600 5300 V5

6400
6276 5200 5163
6200

5100 5086
6000
5784
5800
5000 4970
5600
4900
5400

5200 4800
Huawei 2488H Vender1 Vender2 Vender3 Huawei 2488H V5 Vender1 Vender2 Vender3
V6
SAP BWH Benchmark: 5.2 billion data records, Intel Cooper Lake and Cascade Lake CPU
Source: https://www.sap.com/dmc/exp/2018-benchmark-directory/#/bwh?filters=3782ae74-cbb3-4dc6-ba44-fefe87d9ed4f;6bc6523a-6abc-4c6d-8964-91e7d3b5f5be

29 Huawei Confidential
 Scalable Swift Strong Smart

Huawei SAP HANA Optane Memory TDI Solution(4-socket) Ranks

Top 1 SAP BWH Benchmark
SAP® BWH Benchmark for 4-socket servers
Query Executions per Hour
Top1
HANA
4,760
Benchmark
4,740 4,734
4,729
4,720
Huawei 2488H V6
Optane Memory Soluiton 4,700

4,680

4,660

4,640
4,616
4,620

4,600

4,580

4,560

4,540
Huawei 2488H V6 Vender 1 Vender 2

SAP BWH Benchmark: 7.8 billion data records, Intel Cooper Lake CPU with Intel Optane Persistent memory (1.5T DRAM + 3TB Optane Memory)
Source: https://www.sap.com/dmc/exp/2018-benchmark-directory/#/bwh?filters=v:3782ae74-cbb3-4dc6-ba44-fefe87d9ed4f

30 Huawei Confidential
 Scalable Swift Strong Smart

Huawei Servers Provide the Industry-leading Reliability of x86-

based Universal Servers
Unplanned Downtime Due to Inherent system or component flaws per
Minute/per Server (2018~2020)
Vender 8 42.67

Vender 7 41.67

Vender 6 38.33

Vender 5 29.00

Vender 4 4.30

Vender 3 4.17

Vender 2 2.23
Industry-leading reliability
Huawei 2.17

Vender 1 2.07

IBM Power 1.96

IBM Z 0.75

0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00
Source: ITIC 2018~2020 Global Server Hardware Server OS Reliability Survey Results
https://itic-corp.com/blog/2018/08/itic-2018-server-reliability-mid-year-update-ibm-z-ibm-power-lenovo-system-x-hpe-integrity-superdome-huawei-kunlun-deliver-highest-uptime/
https://itic-corp.com/blog/2019/11/ibm-lenovo-hpe-and-huawei-servers-maintain-top-reliability-rankings-cisco-makes-big-gains-ibm-lenovo-hardware-up-to-24x-more-reliable-28x-more-economical-vs-least-reliable-white-box-servers/
https://itic-corp.com/blog/2020/05/itic-2020-reliability-poll-ibm-lenovo-hpe-huawei-mission-critical-servers-deliver-highest-uptime-availiability/

31 Huawei Confidential
 Scalable Swift Stable Smart

Huawei Provides Diversified HANA DR Solutions, Ensuring

Service Continuity

Fully automated Quasi-real-time Low TCO Easy deployment Abundant

Fewer manual Boost in reliability No need for Professional experience
operation errors third-party implementation DR at a distance of
HA software services over 1300 km

Operation System HA Cluster (Virtual IP)

HANA agent
Jilin
DR center
heartbeat Distance: 1300+ km
monitor Bandwidth: 150 Mbit/s
monitor

Primary Secondary Beijing

(active) (active) Production
System
HANA Replication HANA
center

(Synchronous)
Data Log Data Log
Disks Disks Disks Disks

High availability Geo-redundant DR*

32 Huawei Confidential * Deployment case of top customer in the

petrochemical industry in China
 Scalable Swift Strong Smart

Huawei SAP HANA RAS Enhancement Tool

Real-Time Health Diagnosis for over 80% Higher Reliability
✓ 360-degree diagnosis on large
smart screens
✓ Real-time fault alerts
✓ 80%+ higher system reliability

300+ professional diagnosis indicators, 24/7 real-time

Distribution of Huawei HANA
360-degree health diagnosis for HANA
customers' O&M issues • Single-node health check
4% 1%
intelligent diagnosis • HA cluster health check
6% OS issues • Real-time monitoring
Huawei SAP HANA • DBA expert indicator analysis (such as inappropriate SQL)
21% HANA DB issues • Alerts for other risks (table structure and table data)
RAS enhancement tool
Dual-node HA issues Effective alerts for 80%+ faults, preventing system
Drive issues breakdowns
46% 22% Huawei FusionDirector • Identifies improper SQL statements or processes that will cause HANA
Hardware issues intelligent management exceptions or even system breakdowns
software • Detects abnormal HANA system resource indicators
Storage issues • Detects abnormal synchronization between active and standby HANA
systems or faults that will trigger failover

33 Huawei Confidential RAS: Reliability, Availability, and Serviceability

Huawei One-Stop SAP HANA Solution Meets Customized
Enterprise Database Requirements
Mgmt.
SAP Apps ... ... Platform
SAP Business Suite (OLTP) SAP Business Analysis (OLAP)

SAP
HANA HANA
System SAP System Replication SAP
Replication HANA HANA
Databases
SAP HANA (On-Premise) SAP HANA (Huawei Cloud) SAP System Copy

System
Environment ... ...
s
OS Virtualization

SAP System Clone

Servers

FusionServer Pro & KunLun

External SAP HANA Refresh

Storage BCManager Huawei Storage
OceanStor 5510F Dorado 5000 Dorado 6000 Dorado 18000 eReplication Services Connector
OceanStor/Dorado Storage Software LaMa
34 Huawei Confidential
Huawei SAP HANA Solution Is Your Best Choice

1600+ Honor The Most Comprehensive One-

SAP HANA customers An Honorable Mention in the Special stop Service
Award category of the SAP HANA® 192 GB to 24 TB memory; diversified
Innovation Awards 2017, in recognition SAP solutions, including HANA
of Huawei's continuous innovation appliance, TDI, SAP cloud, and big data

HANA RAS Enhancement Tool

Superior Improve system reliability and simplify
system O&M through OS monitoring,
Performance HANA monitoring, SQL monitoring,
Superior performance of SAP BW system health check and HANA HA
edition for SAP HANA benchmark management

Automated Deployment Tool

1300+ Kilometers 25+ Industries One-click deployment of OS and HANA
Experience of SAP HANA DR Solutions for 25 industries, including
HANA system replication and High
deployment at an ultra-long distance retail, manufacturing, energy,
Availability automatic configuration
technology, and finance

35 Huawei Confidential
Contents
• SAP and SAP HANA Introduction
• Huawei SAP HANA Solution
• Huawei SAP HANA Solution Highlight
• Huawei SAP HANA Configuration and Quotation Guide
• Huawei SAP HANA Resource and Project Case

36 Huawei Confidential
Huawei SAP HANA Solution Configuration using SCT
http://unistar.huawei.com/sct/login!loginChoose.action

2. Input 3. Click
‘SAP HANA’ ‘Search’

1. Click
‘Add Product’

37 Huawei Confidential
First step: Choose the Product

Huawei solutions for

SAP HANA-2288H V5
Huawei solutions for
SAP HANA-2488H
V5/V6
Huawei solutions for
SAP HANA-9008 V5

38 Huawei Confidential
Second Step: Choose Solution (SAP HANA Appliance）
SAP HANA Appliance
(Single Node)

SAP HANA Appliance only uses

top-bin CPUs*

* Top-bin CPUs means Intel two top-level CPUs (including M or L CPUs), for example, Cascade Lake 8276, 8280 and Cooper Lake 8376, 8380.

39 Huawei Confidential
Second Step: Choose Solution (SAP HANA TDI)
SAP HANA TDI
configuration using
external storage

SAP HANA TDI

configuration using
local storage
A wide range of CPUs* are selectable
(including more than 8 cores lower-
end CPUs, not only top-bin CPUs

* Kunlun 9008 only support 8xxx series CPU ， 2288H V5 and 2488H V5/V6 support Bronze, Silver and Gold series CPU

40 Huawei Confidential
Third Step：Input Requirement (1)
Select Single node (Appliance) / TDI (local
storage or external storage) / Scale-out

Select SOH (OLTP) or BWOH (OLAP)

depending on customer's business type

Choose the memory capacity

For low cost, select ‘Base configuration’

Prefer ‘SSD configuration’

Use default configuration

Preferred Cascade Lake CPU

For HANA Appliance, prefers the default CPU;

For TDI, select the CPU flexibly

Prefer the 64GB DIMM, not recommend to

choose 128GB DIMM as high price

For the production system, always choose Yes,

it will automatically add a 10GE Optical network
card

Partition rule for OS:

1. OS + HANA shared in one Raid group；
2. OS independent (prefer)

Select the HANA disk type and size

Allows to increase the number of hard disks，

prefer to use default values
41 Huawei Confidential
Raid card type, prefer to use default setting
Third Step：Input Requirement (2)

According to the actual requirements,

increase PCIe cards, such as network
cards and HBA cards

42 Huawei Confidential
Third Step：Input Requirement (3)

Use default configuration

Select SuSE operating system version based

on customer's actual requirement; Prefer the
latest version

Select license age based on actual

requirement

Default is 1 license support 2 socket or 2

virtual machines, optional 2 socket unlimited
virtual machines

Use default configuration

43 Huawei Confidential
Fourth Step : Configure SAP HANA Service (1)

1. Click
‘Configure Service’

2. Add
‘SAP HANA Solution
Implementaton Service’

3. Click ‘Config’ to go to
the implementation service
configuration page.

44 Huawei Confidential
Fourth Step : Configure SAP HANA Service (2)
1. Enter the number of nodes for which
HANA HA needs to be configured or
2. If Pre-installation is not selected during HANA Scale-out cluster*
BOQ configuration or the pre-installed version
does not meet customer requirements, select
‘Needed’.

3. Enter the number of nodes to be

installed or reinstalled.

* If two SAP HANA production systems need to be configured with HA, enter 2. If you want to install the HANA scale-out cluster in 4+1 mode, enter 4.

45 Huawei Confidential
SAP HANA TDI Configuration with Intel Optane Persistent
Memory
SAP HANA TDI local or external
storage

Choose‘DRAM+Optane PMem’

DRAM type and number

Optane Persistent Memory

（DCPMM）type and number

46 Huawei Confidential
Contents
• SAP and SAP HANA Introduction
• Huawei SAP HANA Solution
• Huawei SAP HANA Solution Highlight
• Huawei SAP HANA Competitive Positioning
• Huawei SAP HANA Configuration and Quotation Guide
• Huawei SAP HANA Resource and Project Case

47 Huawei Confidential
 SAP HANA Certification
SAP HANA Certification URL： https://www.sap.com/dmc/exp/2014-09-02-hana-hardware/enEN/index.html

1 Check ‘Certified Appliances’

2 Select ‘Venders’ 3 Select ‘CPU Architecture’

48 Huawei Confidential
 SAP HANA Benchmark
SAP HANA Benchmark URL： https://www.sap.com/dmc/exp/2018-benchmark-directory/#/bwh

1. The SAP SD benchmark is the performance evaluation index of the SAP application system in traditional non-HANA databases (DB2, ASE, SQL Server,
etc.), which cannot effectively judge the performance of the SAP HANA appliance
https://www.sap.com/dmc/exp/2018-benchmark-directory/#/sd?filters=v:6bc6523a-6abc-4c6d-8964-91e7d3b5f5be;v:24a2ecc0-5986-4eee-9869-
48f755aed644;v:3782ae74-cbb3-4dc6-ba44-fefe87d9ed4f

2. The performance of the SAP HANA appliance is judged by the SAP BWH benchmark. The SAP application system runs on the SAP HANA database
and uses the number of query executions per hour (QPH) of the SAP HANA database as the performance evaluation index; Huawei 2488H V5/V6 BWH
benchmark ranks No. 1；
https://www.sap.com/dmc/exp/2018-benchmark-directory/#/bwh?filters=v:3782ae74-cbb3-4dc6-ba44-fefe87d9ed4f;v:6bc6523a-6abc-4c6d-8964-
91e7d3b5f5be

1 Select ‘Technology Partner’ 2 Select ‘CPU’

49 Huawei Confidential
 Intel® Xeon® V5/V6 CPU for SAP HANA

Intel® Xeon® Cooper Intel® Xeon®

Lake series Cascade lake series SAP HANA TDI CPU: all
SAP HANA Appliance standard CPU: Intel Platinum, Gold and
83xx,63xx,53xx 82xx,62xx,52xx,42xx,32xx Silver CPUs with 8+ cores
1) Cooper Lake 8376H/8376HL/8380H/8380HL
（4S~8S）
2) Cascade Lake 8276/8276L/8280/8280L

Intel® Xeon® Scalable Processors

Platinum Gold Silver Bronze (V5)

Cooper Lake per Cascade Lake per

CPU/RAM ratio CPU/RAM ratio

53xxH/63xxH/83xxH: <=1T 52xx/62xx/82xx: <=1T

63xxHL/83xxHL <=4.5T 52xxL/62xxL/82xxL <=4.5T

50 Huawei Confidential
After this course, you should be able to answer the following questions:

1 Huawei SAP HANA solution supports a maximum of (8)-socket.

2 The maximum (1.5) TB of RAM is supported by a single Intel Cascade Lake or Coopler Lake CPU in
OLTP scenario, for example SAP S/4HANA system.

3 The maximum (0.75) TB of RAM is supported by a single Intel Cascade Lake or Coopler Lake CPU
in OLAP scenario, for example SAP BW/4HANA system.

4 Huawei Kunlun 9008 V5 HANA appliance supports maximum (12) TB RAM. And support maximum
(24) TB RAM is using Intel Optane Persistant memory DIMM.
 SAP HANA Project Case

52 Huawei Confidential
Class Schedule

SN Course Time Course Name (Storage) Trainer

01 2021-03-22, 16:00-18:00 01 Huawei SAP HANA Solution Jiang Hua

02 2021-03-23, 16:00-18:00 02 Huawei vSAN Solution Wang Qi

03 2021-03-24, 16:00-18:00 03 Huawei High Performance Computing Solution Liu Jierui

04 2021-03-25, 16:20-18:00 04 Huawei Computing Management Software FusionDirector Hu Lei

05 2021-03-26, 16:00-17:00 05 Huawei Atlas Enables Industry Solution Chen Ben

06 2021-03-26, 17:00-18:00 06 Huawei Atlas Applications Development Chen Ben

53 Huawei Confidential
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright©2018 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
HCSP-Presales Computing
HUAWEI vSan-solution
 Content

VMware vSAN Introduction

vSAN Configuration Guidance

Huawei vSAN Solution Advantages

2 Huawei Confidential
VMware Introduction
VMware, Inc. is an American publicly traded software company
from California. The company provides cloud computing,
virtualization software and services. One of the first commercially
successful companies to virtualize x86 architectures
Market Share
• IDC Global Cloud System and Service Management Software Market
Share in 2019 Rank 1
• IDC Global HCI Market Share Rank 1
APPs
• IDC Global Virtual Desktop Computing Software Vendor Rank 1
• IDC Global Enterprise Mobility Management Rank 1
• Global IDC unified terminal management software market share Rank 1
Leadership VM OS (Server OS)
• CRN 2020 Best Hybrid Cloud Solution
54% 14% 12% 6% 5% 9%
• CRN 2018 Annual Hybrid Cloud Product Award Windows IBM Red Hat HP/HPE Oracle Others
• Gartner peerinsights, User-selected Best Hyper-converged
Hypervisor
Architecture Software 2019
77% 15% 6%2%
• Gartner Peer Insights 2020 Annual Customer Selection Award (IT
VMware Microsoft Citrix Others
Infrastructure Monitoring Tool)
• SC Awards 2019 Best mobile security solution Hardware Layer
• Gartner Peer Insight Customer Selection Award: Best Endpoint 16% 16% 9% 7% 6% 5%
HPE/New H3C Dell IBM Inspur Lenovo Huawei
Protection Platform 2019

3 Huawei Confidential
VMware SDDC Product Introduction
• Request resources through the user interface or API and
Cloud orchestrate lower-layer actions.
Automation • Cloud automation layer configuration is part of SDDC cloud
operation and automation solution implementation.
Layer

• Controls access to underlying physical infrastructure and

allocates resources to management and tenant workloads.
Virtual Infra • Virtual infrastructure layer configuration is part of the
Layer SDDC management domain and workload domain
implementation.

• Computing, network, and storage components. Computing

components include x86 servers, running management
Physical Infra components, NSX-T edge nodes, and tenant workloads.
Layer • The physical infrastructure layer configuration is part of the
SDDC management domain and workload domain
implementation.

Veeam
Finance, government, healthcare, transportation...
1. VMware SDDC consists of the cloud automation layer, virtualization
FD for vCenter architecture layer, and physical architecture layer.
Plugin
2. In the SDDC infrastructure, Huawei provides server hardware, server
Horizon vRealize Suite
Stretched management software, and vCenter plug-in.
cluster
3. The red part is the part that Huawei can participate in.
VAAI vSphere NSX Tanzu Pivotal SRM
Huawei Products

FusionServer hyper-converged infrastructure VMware Products

vSAN
（X86 Servers）
Third-parties Products

4 Huawei Confidential
Business Scenarios Insight for vSAN

User Application Challenges and Requirements Reason for selecting vSAN Customer Cases
Scenarios Scenarios
• Complex management for traditional storage
• Simple Management
• Dev & Test System • Reduced storage investment
A Dev & Test
• DevCloud • Frequent system changes
• No need to purchase high-end storage • British Army
• linear expansion
• Good scalability

• New application rollout • Reduces device costs and saves energy

• New service system performance assurance
• Web Applications • Improve operational efficiency
Production • Reduces risks brought by infrastructure faults to • Temple University
B Application
• Database
applications
• Greatly improve performance
• APPs • Greatly reduces the TCO
• Flexible application performance adjustment
• Big Data MapReduce • Storage virtualization improves utilization

• High cost-effectiveness
• Reduces the overall cost of VDI
• Virtual Desktop • No large investment in external disk arrays • National Bank of
C VDI • Improved user experience and performance
• Desktop Cloud • Online expansion with the VDI scale Georgia
• Simplify storage management
• Simple management

• High-performance I/O • High I/O throughput and low latency

• Private Cloud
• Massive storage pool • Large capacity and easy expansion
Cloud/ • City Cloud
D • High scalability • High cost-effectiveness • Baystate Health
SDDC • Industry Cloud
• Cost pressure • Integrated with the cloud platform for
• Resource Pool
• Automatic deployment automatic deployment

5 Huawei Confidential
Virtual SAN Architecture and Usage Requirements
vSphere Version Requirement
• vSphere 5.5 Update 1 or higher
Hardware Requirements
• All hardware of Virtual SAN must be certified and published on VMware HCL
• I/O Controller, HDD and SSD Must be certified and published on by vSAN HCL
• Using uncertified hardware can lead to performance problems and/or data loss because
vSphere
the behavior of uncertified hardware is unpredictable.
• VMware cannot provide technical support for environments running on uncertified
hardware
vSAN Datastore
• Capacity Tier – Capacity • Disk Group – Performance and
Capacity
• Cache Tier – Performance
• Node – Compute Capability
Hybrid All Flash
disk group disk group disk group disk group disk group

SSD PCIe Ultra DIMM

SSD PCIe Ultra DIMM
The written content is cached first,
scale up

Read and Write Cache and the read content is directly

scale out Cache Tier transferred to the capacity tier.
HDD HDD HDD HDD HDD

Capacity
Capacity tier Capacity Tier Flash Device
VSAN network VSAN network VSAN network VSAN network SAS/NL SAS/SATA/Passthrough JBOD The read content goes to the capacity tier
Per Host 40K IOPS Per Host 90K IOPS + Sub-millisecond
Latency

6 Huawei Confidential
Virtual SAN Limitations
Max Configuration version 7 version 6.7
vSAN ESXi host
vSAN disk groups per host 5 5
SSD disks per disk group 1 1
Spinning disks in all diskgroups per host 35 35
Components per vSAN host 9000 9000
Cache tier maximum devices per host 5 5
Capacity tier maximum devices per diskgroup 7 7
Capacity tier maximum devices 35 35
vSAN Cluster
Number of vSAN hosts in a cluster 64 (Hybrid) 64 (All-Flash) 64 (Hybrid) 64 (All-Flash)
Number of datastores per cluster 1 1
In vSAN Stretched Cluster maximum number of data hosts on each site is 15. 30 (15 on each site x 2) 30 (15 on each site x 2)
Maximum number of witness hosts in a vSAN Stretched Cluster or 2-Node Direct Connect configuration 1 1
Maximum number of data hosts in a vSAN 2-Node Direct Connect configuration 2 2
vSAN virtual machines
Virtual machines per host 200 200
Virtual machines per cluster 6400 6400
Virtual machine virtual disk size 62 TB 62
Disk stripes per object 12 12
Percentage of flash read cache reservation 100 100
Percentage of object space reservation 100 100
vSAN networks/physical network fabrics 2 2
vSAN iSCSI Target
Number of iSCSI LUNs per Cluster 1024 1024
Number of iSCSI Targets per Cluster 128 128
Number of iSCSI LUNs per Target 256 256
Max iSCSI LUN size 62 TB 62TB
Number of iSCSI sessions per Node 128 128
iSCSI IO queue depth per Node 4096 4096
Number of outstanding writes per iSCSI LUN 128 128
Number of outstanding IOs per iSCSI LUN 256 256
Number of initiators who register PR key for a iSCSI LUN 64 64

7 Huawei Confidential
Virtual SAN Hardware Requirements
Storage Devices Requirements
• All capacity devices, drivers, and firmware must be certified and listed on the VMware Compatibility Guide List
Cache Tier
• One and only one SAS/SATA SSD or PCIe Flash per disk group
• In Hybrid configurations, cache disk size must be greater than 10% of the capacity size
• In ALL-FLASH configurations, please refer to the reference on the right (normally 5% to 10%)
Capacity Tier
• In Hybrid configurations, at least one SAS or NL-SAS HDD per disk group
• In ALL-flash configurations, at least one SAS or SATA SSD, or PCIe Flash device
I/O Controller
• At least 1 SAS/SATA HBA card, or Raid Controller（Pass-through mode or RAID 0）
• It is NOT recommended to use one I/O controller for both vSAN and non-vSAN storage devices. If vSAN disks work in RAID mode, non-vSAN disks must also work
in RAID mode.
• The VSAN requires that the queue depth of the RAID controller card be greater than or equal to 256
• Currently, VMware does not support SAS expander. When the number of disks on an ESXi host exceeds the maximum number of disks supported by a disk
controller, an additional disk controller is required
Memory
• The memory requirements of vSAN depend on the number of disk groups and devices that the ESXi system management program must manage. For more
information please refer to https://kb.vmware.com/s/article/2113954
Boot Device
• Huawei servers support the dual M.2 SSD RAID 1 solution. Configure RAID 1 for local disks to install the OS. The local disks for installing the OS and the VSAN disk
group cannot use the same RAID controller card at the same time. An extra RAID controller card is required
8 Huawei Confidential
Virtual SAN Other Requirements
Cluster Requirements
• A standard vSAN cluster requires at least three hosts to provide capacity for the cluster. A two-host vSAN cluster
consists of two data hosts and one external witness host
• Hosts in a vSAN cluster cannot participate in other clusters.
Software Requirement
• To use the full vSAN features, the ESXi hosts participating in the vSAN cluster must be 6.7 Update 3 or later vSphere+vSAN Required Licenses
Network Requirements
• Each host must have a minimum bandwidth dedicated to vSAN: 1 Gbit/s dedicated for Hybrid configuration. All-
1. vSphere License
flash configuration dedicated or shared 10 Gbit/s
2. vCenter Server License
• Connection between hosts: Each host in a vSAN cluster, regardless of capacity, must be configured with a
VMkernel network adapter for vSAN communication 3. vSAN License
• Host network: All hosts in the vSAN cluster must be connected to the vSAN Layer 2 or Layer 3 network
• Support for IPv4 and IPv6: vSAN supports both IPv4 and 6
• Network latency: The maximum RTT latency between all hosts in the same cluster is 1 ms. The maximum RTT
between two extended clusters is 5 ms. The maximum RTT between the vSAN witness server and the primary
site is 200 ms.
License Requirement
• vSAN License
• You can allocate a standard vSAN license or a license containing advanced functions to a cluster. Advanced
features include RAID 5/6 erasure coding, deduplication, and compression
• To support encryption and stretched cluster, enterprise-level licenses are requirede

9 Huawei Confidential
Virtual SAN Capacity Planning

Raw Capacity
• To determine the raw capacity of the Virtual SAN data store, multiply the total number of disk groups in the cluster by the size of the capacity disks in those disk groups, and subtract the
overhead for the Virtual SAN disk format
PFTT(Primary level of Failures to Tolerate)
• The fault tolerance mechanism must be considered during capacity planning. The main fault tolerance mechanisms of vSAN includ e Raid-1 (Mirroring) and Raid-5/6 (Erasure Coding)

• X indicates the usable capacity

• Deduplication and compression are required in all-flash scenarios. The deduplication and
compression ratio is related to workloads and is generally more than twice of the workload
• 30% redundant storage space is recommended to avoid data space reconstruction triggered
when the storage usage reaches 80%
• Erasure codes are not supported in hybrid configurations

10 Huawei Confidential
Virtual SAN Sizing Tool
URL：https://vsansizer.vmware.com/ 1. Select the Configuration type 3. Configure the cluster setting
2. Select the vSAN version 4. View the sizing result

Usable Capacity

Raw Capacity

Node Level Configuration

11 Huawei Confidential
Virtual SAN License Introduction

vSAN license Versions

• Standard
• Advanced
• Enterprise
• ROBO （Standard, Advanced, Enterprise）

• Note: Huawei does not resell vSAN temporarily. Customers need

to purchase vSAN by themselves or cooperate with partners and
VMware resellers to provide software services.
• Note: The vSAN license does not include the VMware vSphere
and VMware vCloud licenses and cannot be converted or
upgraded to other types of licenses. Additional purchases are
required if necessary
• For more information please refer to：
https://www.vmware.com/content/dam/digitalmarketing/vmware/en
/pdf/products/vsan/vmware-vsan-66-licensing-guide.pdf
• https://www.vmware.com/reusable_content/vsphere_pricing.html

12 Huawei Confidential
Virtual SAN ROBO （ Remote Office Branch Office ）
Pain Points
• Lack of IT staff in branches
• Inconsistent host configurations
• Limited IT budget
• Limited physical space

vSAN ROBO advantages

• Low license investment
• Charging by VM, with 25 VM licenses in one ROBO license package
• Multiple sites can share the license

vSAN ROBO Limitation

• A maximum of 25 VMs can be deployed at all sites. If more than 25 VMs are deployed,
you need to convert the licenses to other types (Standard, Advanced, and Enterprise)
and charge by CPU
vSAN ROBO Infrastructure
• Dual-host architecture for branches
• A Witness witness server is deployed for each ROBO site at the primary site. No license
is required for the witness server

13 Huawei Confidential
VMware Partnership
GSI & Advisory Partners：

Technology partners and independent software vendors (ISVs) ：

Global Partner Enquiry Website ：https://partnerlocator.vmware.com/?lang=cn#sort=relevancy

14 Huawei Confidential
 Content

VMware vSAN Introduction

vSAN Configuration Guidance

Huawei vSAN Solution Advantages

15 Huawei Confidential
Virtual SAN Solution Types——ReadyNodes
Definition of Ready Nodes：
• vSAN Ready Nodes are x86 servers provided by server vendors that have been pre-configured, tested, and certified for VMware hyper-converged infrastructure
software
• Each Ready Node is optimally configured for vSAN and has the required CPU, memory, network, I/O controller, and storage capacity (SSD, HDD, or flash device)
• Ready Nodes supports only conditional BOM adjustment

vSAN ReadyNode BOM allowed changes vSAN ReadyNode BOM allowed changes for storage

16 Huawei Confidential
Virtual SAN Ready Nodes Configuration
To search for vSAN Ready Nodes information please refer to the link：https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vsan

1. Select the “vSAN” section in the drop-down box.

2. Choose the Ready Node type (All Flash or Hybrid)
3. Choose the Ready Nodes version
4. Choose the Ready Node vendors
5. Choose the CPU type
6. Choose the type: HY refers to Hybrid, AF refers to
All Flash
7. Choose the server type
8. Enter the key word
9. Click update to view the results

17 Huawei Confidential
Virtual SAN Ready Nodes Configuration
The new Ready Node certifications Huawei has finished（URL：
https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vsan&details=1&vsan_type=vsanreadynode&vsan_partner=242&vsan_releas
es=508&page=1&display_interval=10&sortColumn=Partner&sortOrder=Asc）and the roadmap
1. New certification solutions are being tested and
released
2. New certification is driven by project requirements and
no separate roadmap is set

Note: The cooperation with VMware was resumed

in June 2020. The cooperation mainly involves
software and hardware mutual certification,
excluding software resale
Note: The mainstream model for sale is 2288H V5
The roadmap URL for ReadyNode certification:
http://3ms.huawei.com/documents/docinfo/49609
1942647287808

18 Huawei Confidential
Virtual SAN Ready Nodes Typical Configurations
vSAN ReadyNode Profiles
AF-8 Series AF-6 Series AF-4 Series
Raw Storage Capacity per Node 12 TB 8 TB 4 TB
CPU Core 24 core 24 core 20 core
Memory 384 GB 256 GB 128 GB
Capacity Tier Flash 12 TB (Minimum 3 drives in each disk 8 TB (Minimum 3 drives in each disk 4 TB (Minimum 2 drives in each disk
group) group) group)
SSD Endurance Class A or above SSD Endurance Class A or above SSD Endurance Class A or above
SSD Performance Class C or above SSD Performance Class C or above SSD Performance Class C or above
Caching Tier Flash 2x400 GB 2x200 GB 1x200 GB
SSD Endurance Class D or above SSD Endurance Class C or above SSD Endurance Class C or above
SSD Performance Class E or above SSD Performance Class D or above SSD Performance Class C or above
IO Controller Queue Depth >= 512 Queue Depth >=512 Queue Depth >= 256
NIC 10 GbE or more 10 GbE or more 10 GbE or more

vSAN Hybrid Hardware Guidance

vSAN ReadyNode Profiles
HY-8 Series HY-6 Series HY-4 Series HY-2 Series
Raw Storage Capacity per Node 12 TB 8 TB 4 TB 2 TB
CPU Core 24 core 20 core 16 core 6 core
Memory 384 GB 256 GB 128 GB 32 GB
Capacity Tier Disk 12 TB (Minimum 3 drives in each 8 TB (Minimum 3 drives in each 4 TB (Minimum 2 drives in each 2 TB (Minimum 2 drives in each
disk group) disk group) disk group) disk group)
SAS 10K RPM NL-SAS 7.2K RPM NL-SAS 7.2K RPM NL-SAS 7.2K RPM
Caching Tier Flash 2x400 GB 2x200 GB 1x200 GB 1x200 GB
SSD Endurance Class >=D SSD Endurance Class >=C SSD Endurance Class >=C SSD Endurance Class >=B
SSD Performance Class >=E SSD Performance Class >=D SSD Performance Class >=D SSD Performance Class >=B
IO Controller Queue Depth >= 512 Queue Depth >= 256 Queue Depth >=256 Queue Depth >=256
NIC 10 GbE 10 GbE 10 GbE 10 GbE

19 Huawei Confidential 查询网址：https://www.vmware.com/resources/compatibility/vsan_profile.html

Virtual SAN Solution Types——Build Your Own
Definition of Build Your Own Configuration
• Based on certified parts
• Main parts（I/O Controller、HDD、SSD）need to be certified by vSAN（URL：https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vsanio）
• Server/CPU needs to be certified by Server Certification, Other parts (NIC) need to be certified by IOVP certification

Server IOVP
Part vSAN IO ReadyNode
Certification Certification
Server/CPU Yes No No Yes
Raid No Yes Yes Yes
SSD/NVMe No Yes Yes Yes
HDD/SSD No Yes Yes Yes
NIC No Yes No Yes
USB No Yes No Yes

20 Huawei Confidential
Virtual SAN Solution Configuration Process（BYO）
1. Check the OS Compatibility of the Server（URL：https://www.vmware.com/resources/compatibility/search.php?deviceCategory=server ）

1. Choose the System/Server type

2. Choose the vSphere Version
3. Select the vendor Huawei
4. Input the keyword of the server model
5. Click and search for the result

Vendor Information Server Model Supported Version

CPU Series

21 Huawei Confidential
Virtual SAN Solution Configuration Process（BYO）
2. Check the compatibility of the IO Devices (Including NIC, FC and etc.) （URL： https://support-it.huawei.com/ftca/zh/product/rack-server ）

1. Choose the IO Device type (IOVP

Certification)
2. Choose the vSphere Version
3. Choose the I/O device type
4. Select the vendor Huawei
5. Input the keyword of the NIC
6. Click and search for the result

Vendor Information Model Info Device Type Supported Version

22 Huawei Confidential
Virtual SAN Solution Configuration Process（BYO）
3. Check if the I/O Controller, SSD, HDD has been certified by vSAN Certification
（URL：https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vsan ）

Firstly, click the Build Your Own based on Certified Components

hyperlink in the vSAN page
1. Select the component type
2. Select the vendor
Vendor Information
3. Select the certified version
4. Input the keyword of the component
5. Click and search for the result

Vendor Information Model Info

Endurance Info
Certified Tier
Endurance Class: A-F Supported vSAN version
Performance Class: A-F

23 Huawei Confidential
Practice Case 1——Framework requirement
Requirements

- Total Memory RAM: 1,5 Tb.

- Total Cores Physical: 200 Cores 2.1GHz
- Total Usable Storage: 100TB (SSD) without deduplication and compression.
- Warranty Mission critical 7x24x4 by 3 years.
- Hypervisor: VMware +vSAN (will be provide by partner)

General design idea:

1. Confirm the usage scenario to determine the CPU level (SILVER, GOLD, or PLATINUM). The CPU level corresponds to low-end (ROBO), mid-range (VDI), and high-end
(core applications and databases)
2. Calculate the total number of required physical cores (the ratio of physical cores to virtual cores is generally 1:3 or 1:4; the ratio of physical cores to virtual cores is
generally less than 1:6) and estimate the number of servers (Number of servers = Total number of physical cores/Number of cores per CPU/2).
3. Calculate the physical memory size (the ratio of physical memory to virtual memory is generally 1:1, and the ratio of physical memory to virtual memory is less than or
equal to 1:1.5 when a disaster occurs), and evenly allocate the physical memory size to each node based on the number of servers calculated in step 2
4. Calculate the raw capacity of the cluster (the minimum raw capacity is FTT=1 and RAID5/6), and allocate the servers calculated in step 2 to all nodes evenly
5. Configure a single node, select the disk type, and estimate the number of disks
6. Calculate the Cache Disk Capacity Based on the Capacity Disk Capacity
7. Configure the other parts including NIC, RAID, OS Boot Device and etc.
Note: Generally, 15% to 30% redundancy is provided for the memory, number of cores, and storage space required by customers t o ensure stable system running

24 Huawei Confidential
Practice Case 2——Requirements based on VM
TOTAL
CPU
Requirements Servers cores Memory Capacity Cache
TYPE1 (12 sites) 4 156 1280 30 912
TYPE2 MAZOWIECKIE 18 623 5760 274 8208
TYPE1 SLASKIE 6 236 1920 46 1368
TYPE1 MALOPOLSKIE 5 183 1600 38 1140
TYPE2
WIELKOPOLSKIE 11 412 3520 167 5016
TEST A 16 384 1024 28.8 0
TEST B 2 96 512 6 0
General design idea:
1. Confirm the virtual cluster design solution and classify configuration types based on cluster application scenarios
2. Design node configurations for each cluster, summarize cluster VM requirements, and generate an overall resource pool
3. Confirm the usage scenario to determine the CPU level (SILVER, GOLD, or PLATINUM). The CPU level corresponds to low-end (ROBO), mid-range (VDI), and high-end
(core applications and databases)
4. Calculate the total number of required physical cores (the ratio of physical cores to virtual cores is generally 1:3 or 1:4; the ratio of physical cores to virtual cores is
generally less than 1:6) and estimate the number of servers (Number of servers = Total number of physical cores/Number of cores per CPU/2).
5. Calculate the physical memory size (the ratio of physical memory to virtual memory is generally 1:1, and the ratio of physical memory to virtual memory is less than or
equal to 1:1.5 when a disaster occurs), and evenly allocate the physical memory size to each node based on the number of servers calculated in step 2
6. Calculate the raw capacity of the cluster (the minimum raw capacity is FTT=1 and RAID5/6), and allocate the servers calculated in step 2 to all nodes evenly
7. Configure a single node, select the disk type, and estimate the number of disks
8. Calculate the Cache Disk Capacity Based on the Capacity Disk Capacity
9. Configure the other parts including NIC, RAID, OS Boot Device and etc.
Note: Generally, 15% to 30% redundancy is provided for the memory, number of cores, and storage space required by customers t o ensure stable system running

25 Huawei Confidential
Practice Case 3——Based on Clear Configuration Requirements
Requirements

Base Configuration General design idea:

· Must be rackmount and All flash vSAN ReadyNode certified and verified by VMware
· Must support 24 DDR4 DIMM slots, Supports RDIMM /LRDIMM, support up to 12 NVDIMM, Supports registered ECC DDR4 DIMMs only 1. Select the configuration framework based on the
· Front bays: Up to 24 x 2.5” SAS/SATA (HDD/SSD), NVMe SSD max 153TB
· PCIe must have minimum feature support as DPDK, iSCSI offload, NVMe-oF and NVME-oEthernet, Jumpo Frame, Tunnel offload support for VXLAN, customer's configuration parameters, including the
NVGRE, and GENEVE.
· Must support following integrations: Microsoft® System Center, VMware® vCenter™, Nagios & Nagios XI, HP Operations Manager, CA Network and model, CPU, memory, NIC, and RAID controller
Systems Management, ServiceNow and Dell EMC OpenManager.
· Must support TPM 1.0/2.0, Secure boot, Secure Erase, Cryptographically signed firmware and system lockdown.
card.)
· 1GE out of band management network
2. Plan the cluster scale and resource pool based on
· Pre-Installed ESXi and it must support ESXI 6.7 U2/U3 and ESXI 7.0
§ BOSS card with dual M.2 SSD each with 240GB protected by RAID 1 for ESXI the usable capacity
Processor Configuration
· Intel Xeon Gold 6230 (2.1GHz/16-Core) Cascade lake Processor(with heatsink) 3. Determine the single-node configuration based on
Memory Configuration
· Total of 1.5TB RAM (24 x 64 GB RAM) the cluster scale and resource pool capacity
Disk Configuration
· Disk controllers must be listed in VMware compatibility guide 4. Configure a single node based on the configuration
· Minimum of two disk groups
· Minimum of 60-100 TB Usable capacity (RAID-5 and FFT-2 when clustered with 8 nodes cluster) framework confirmed in step 1. For details, see
· Cache Tier must have:
§ Endurance: 35040 TBW cases 1 and 2
§ Endurance Class: Endurance Class D >=7300 TBW
§ Performance Class: Class E: 30,000-100,000 writes per second
· Capacity Tier must have: Note: Generally, 15% to 30% redundancy is provided
§ Endurance: Minimum of 3650 TBW
§ Endurance Class: Minimum Endurance Class D >=3650 TBW for the memory, number of cores, and storage space
§ Performance Class: Class E: 30,000-100,000 writes per second
PCIe Card NIC required by customers to ensure stable system running
· 2 x dual port 10/25GB with Bi-directional SFP connectivity.
PCIe Card-FC card
· 1 x dual port 16/32GB with Bi-directional SFP connectivity.

26 Huawei Confidential
 Content

VMware vSAN Introduction

vSAN Configuration Guidance

Huawei vSAN Solution Advantages

27 Huawei Confidential
Huawei Self-developed High-Performance NVMe SSD
VSAN IOPS performance > 2,000,000
• Pain point: The traditional HDD+SSD hybrid vSAN architecture has low
performance and cannot meet new service requirements. 10X faster
• Analysis: Low-speed HDDs severely affect I/O performance
• Solution: All NVMe SSD Solution
• Benefit: 10x Performance Improvement 160,000

4 * Hybrid vSAN servers 3 * All NVMe SSD vSAN servers

• >150 Core Technology Patents Huawei SSD Controller

Chip
• >10 years Development Experience
• The first 800k IOPS
• The first to support FW activation online
+
4 K R N D R EAD IOPS K /S 4 K R N D W R ITE IOPS K /S
Intel P4610 Samsung PM1725b Huawei ES3500P V5 Intel P4610 Samsung PM1725b Huawei ES3600P V5

825
825
800

800

300
740
720

640

640

250
620

220
200

200
195

190
180
135
1.6TB 3.2TB 6.4TB 1.6TB 3.2TB 6.4TB
28 Huawei Confidential
Huawei M.2 SSD - Hot-swappable and RAID 1 Supported
• Pain point: Traditional ESXi installation using a single disk or SD card cannot
meet customers' new maintenance requirements. Otherwise, even raid 1 is
High-speed and Stable M.2 SSD
enabled for the boot devices, server has to be powered off to replace the faulty
• Independent OS installation and high-speed
part startup
• Analysis: ESXi Single-Disk Deployment Affects System Reliability. Traditional • Hot-swappable, Hardware RAID, High
Reliability
dual disk deployment are not hot-swappable
• Solution: M.2 that supports hot-swappable and hardware RAID 1 is used to
install and deploy ESXi on the built-in storage.
• Benefit: Maintenance personnel can directly replace the faulty part on site

Other Vendor Huawei M.2 SSD

Deployment
Or
+

VS
• For single disk deployment, if the disk gets faulty, the server goes down • In Huawei M.2 raid 1 deployment, even one disk gets faulty, the server
too will not be affected.
• For dual disk deployment, server has to be powered off to change the • The faulty part can be replaced directly without powering off the server.
faulty disk 0 interruption for bussiness
29 Huawei Confidential
FD for vCenter——Unified Management Plug-in
Huawei FD for vCenter Plug-in —— Unified and integrated management portal on vCenter

vCenter Web Client

FDIVV
Out of Band Management

FusionDirector
Unified Management

Server 1 Server 2

iBMC iBMC
Out-of-Band Management Out-of-Band Management Supported Functions:
Redfish Protocol Redfish Protocol
• Server information display, including event, warning and error message
OS OS
• Server deployment(BIOS、RAID、NTP、DNS、LDAP)
• OS deployment
• Firmware and Driver update
• Proactive HA

30 Huawei Confidential
2288H V5 Server Breaks VMware VMmark Benchmark Test Records
Benchmark test with the same workload, with higher performance than
competitors' products

28 Tiles Vmmark Performance Test 8 Tiles Vmmark Performance Test

+12% +7%

Vmmark Performance Test Query URL：https://www.vmware.com/products/vmmark/results3x.0.html?sort=date&storage=vSAN%20Storage

VMmark is a performance benchmark test launched by VMware. It simulates

actual services, such as the email server, Java server, and backup, to measure
the overall system performance. The service pressure is measured by tile.

31 Huawei Confidential
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright©2018 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
HCSP-Presales Computing
Huawei High Performance computing solution

1
Content

1 HPC introduction

2 HPC solution decomposition

3 Success case

2
Definition

High Performance Computing generally refers to the practice of

aggregating computing power in a way that delivers much higher

performance than one could get out of a typical desktop

computer or workstation in order to solve large problems in science,
engineering, or business.

3
HPC Applications

Satellite Mapping Meteorological Science Energy Exploration Aerospace

Life Sciences Basic Science Research Automotive Electronics Molecular Dynamics

4
Huawei HPC Solution Architecture

L4
Service Solution
Energy Manufacturing Gene sequencing Weather
applications
design

Math
MPI computing Compiler
Training&
communication library Enablement

L3
Cluster Job
Operating System
management scheduling
system deployment

L2
Application
Server Storage Network tuning

Joint

L1 Infrastructure Data center Liquid

cooling
inovation

5
HPC Cluster Infrastructure

• Compute
• Compute node
• Fat node
• GPU accelerated node

• Storage
• Parallel file system – Lustre, GPFS
• Distributed file system

• Interconnection
• IB：InfiniBand, dominated by Mellanox (acquired by NVIDIA)
• RoCE

• Management & Utility nodes

• Management
• Login / Gateway
• Pre/Post processing

Cluster composition Compute Storage Interconnection

Theoretical Double precision float point Raw capacity (PB) and Bandwidth (Gbps) and Latency
performance Indicator operation per second (FLOPs) aggregation bandwidth (GB/s) (nanosecond)
Achievable Application performance Usable capacity and tested Tested results
Performance benchmark / Linpack / HPCG bandwidth

6
Evaluation method of computing capability
The HPC cluster is measured in Floating-point Operation Per Second (FLOPS)

Theoretical peak performance (Rpeak) Maximal achieved performance (Rmax)

Rmax is scored based on their performance
using the HPL / HPCG Benchmark.
Double precision FLOPS
Intel V2 CPU (Sandy Bridge and Ivy Bridge) = 8 FLOPs/cycle High performance Linpack High Performance Conjugate
Intel V3/V4 CPU (Haswell, Broadwell) = 16 FLOPs/cycle (HPL) Gradients (HPCG)
Intel Scalable Processor CPU (with 2 FMA) = 32 FLOPs/cycle
- A dense n by n system of linear - a complement to HPL.
2-socket server configured with Scalable equations Ax = b. - designed to exercise
computational and data access
Processor 6248R (24 core, 3.0 GHz) - Application performance is patterns
much lower than HPL. - match a broad set of applications
GFLOPS per node = 2 x 24 x 3.0 x 32 = 4608 - HPCG list
- TOP500 list.

𝑅𝑚𝑎𝑥
System efficiency =
𝑅𝑝𝑒𝑎𝑘 Efficiency 60% ~ 80% 0.5% ~ 10%

7
Content

1 HPC introduction

2 HPC solution decomposition

3 Success case

8
HPC workflow
Workflow

Compute node Compute

Cluster manager Storage

Job scheduling
4 Acceleration Workstation Interconnect
Fat node
(GPU) (optional)
Summary
Management node Login node
3 2 5 1

Management Application installation,

Interconnection
network 1 upload raw data

Submit jobs script to job

2 scheduler
Storage IO node
Scheduler allocates workload to
Storage
system
3 computing cluster
Application
Obtain data from storage and
MPI 4 start calculation
Compiler
Math library User check job status and fetch
5 final results from storage

9
Working with Intel to Build Flexible, Intelligent,
and Premium FusionServer Pro Intelligent Servers Workflow
2nd Intel Xeon Scalable Processor
Compute
Mission-critical server Full Liquid Cooling Solution E9000 Blade server
Storage
Liquid-cooled
General Computing-
computing GPU expansion All-NVMe storage
computing intensive
KunLun 9008 V5 Interconnect
X6000 XH321L V5 E9000 CH121L V5 CH121L V5 CH121 V5 CH242 V5 CH221 V5 CH225 V5 Summary
Rack server High-density server
General Storage- Computing- Key service Storage-
Balanced server type 2 U 4 nodes Computing-intensive 4 U 4-node
computing intensive intensive intensive

1288H V5 2288/H V5 5288 V5 2488H V5/V6 5885H V5 X6000 XH321 V5 XH321L V5 X6800 XH628 V5

Intelligent acceleration engine Intelligent management engine

AI
SSD iNIC Accelerator
FusionDirector
Card
Five Intelligent Management
Features
ES3000 V6 IN200/300/500 Atlas 300 Nvidia A100 Nvidia V100

Nimble Intelligent High quality

Adapt to any business needs Intelligent upgrade, reducing OPEX by 15% The fault rate is 15% lower
than the industry average.
• Various rack, blade, and high-density models • Five intelligent • Selected components and derating design
• Mission-critical servers and full liquid management features • 10-12 Signal BER, strict test, and global
cooling solution • Three intelligent
certification
acceleration components

Serving 10,000+ industry customers worldwide, covering 1/3 of the world's population

10
Main CPU and GPU policies in HPC scenarios:
Avoid bidding by core count for CPUs. Promote PCIe A100 for GPUs Workflow

1. AMD use multiple cores, the performance of a single core is 1. The FP64 floating-point performance of the NV-LINK is the Compute
weak. same as that of the PCIe A100.
2. HPC focuses on the FP64 floating-point performance. Persuade 2. The main difference lies in the memory bandwidth. NV-LINK is Storage
customers not to use the core count as a cluster indicator. suitable for AI scenarios. In HPC scenarios, persuade customers
to use PCIe A100, which also has lower power consumption.
Interconnect
Intel Icelake CPU AMD Rome CPU
NVIDIA Summary
SKU 8358 7H12
SKU A100 NVLINK A100 PCIe
Frequency（Ghz） 2.6 2.6
FP64 Rpeak 9.7 9.7
Core count 32 64 （TFLOPS）
GPU vRAM 40GB/80GB 40GB
Float operation per 32 16
cycle GPU memory 1,555 GB/s or 1,555 GB/s
FP64（TFLOPS） 2.662 2.662 bandwidth 2,039 GB/s
TDP（W） 400 250
TDP（W） 250 280
Form NV-LINK PCIE 4.0

https://developer.nvidia.com/hpc-application-performance

11 Huawei Confidential
Huawei HPC Storage Solution
Workflow

Type 1 Type 2 Type 3

Compute

NAS storage OceanStor Pacific Lustre parallel file system Storage

distributed file system
Compute cluster Compute cluster Compute cluster Interconnect

… … … Summary

10GE IB IB/10GE

OceanStor
Pacific
MDS (X86)
…
Lustre file system
OSS (X86)

…
OceanStor NAS OceanStor 100D

For light weighted cluster Huawei dedicated solution Typical configuration

12
Storage: OceanStor Pacific, defining next-generation HPDA storage
Workflow

Compute

Application Autonomous oil and gas Meteorologi For hybrid loads - higher Storage
EDA AI Life Sciences
driving exploration cal Ocean
performance
Interconnect
Posix MPI One storage system supports high bandwidth and
high OPS, 32 GB/s bandwidth and 400,000 IOPS/U,
NFS DPC Summary
Protocol CIFS HDFS S3 leading in performance density.
(Distributed Parallel Client)

Ultra-high-density design -
Load balancing Quota Grading Metadata retrieval
lower cost
Management
Software 24 disk slots/U high-density and large-capacity
Snapshot Remote replication Multi-tenant QoS system
hardware, intelligent tiering of hot and cold data,
OceanFS Parallel File System balancing data value and cost

Multi-protocol interoperability -
Product better efficiency
Lossless interworking between
Performance Capacity-based POSIX/NFS/CIFS/HDFS/S3, zero data copy, and
evolution towards HPDA and AI

OceanStor Pacific Storage

13
Full-series dedicated storage, meeting requirements in different scenarios
Workflow

Performance Capacity Compute

Storage

ultra-high performance orientated Large-capacity warm and cold Interconnect

such as AI training and EDA and data scenarios, such as oil and Summary
other hot data scenarios gas exploration and
5 U 8-node all-flash/chassis
autonomous driving
1 x Kunpeng 920 CPU/node
5 U 2-node HDD/enclosure
10 x NVMe SSDs per node
1 x Kunpeng 920 CPU/node
OceanStor Pacific 9950 Up to 160 GB/s bandwidth, 2 million IOPS per OceanStor Pacific 9550
60 x 7.2K RPM SATA/node
chassis
Maximum 10 GB/s bandwidth per chassis

Small-scale and common Medium-capacity warm and

performance scenarios cold data scenarios
2 U all-flash node 4 U HDD node
2 * Kunpeng 920 CPU 2 * Kunpeng 920 CPU
12 * SAS SSD 36 * 7.2K RPM SATA
OceanStor Pacific 9920 OceanStor Pacific 9540
Up to 9 GB/s bandwidth and 140,000 IOPS Maximum bandwidth: 3.6 GB/s

14
High-speed interconnection network:
IB and RoCE high-speed network Workflow

Compute

Storage

Interconnect

Summary

InfiniBand Technology Evolution:

10 Gbit/s SDR 20 Gbit/s DDR 40 Gbit/s QDR 56 Gbit/s FDR 100 Gbit/s EDR 200 Gbit/s HDR 400 Gbit/s NDR XDR

...

Adapter High-speed Switch Software Supports multiple network topologies

cable

15
HDR + X6000 design reference
Workflow
Spine switches (L2) :
2 * 40-port InfiniBand HDR managed switch QM7800
Leaf switches (L1) :
HDR 200Gbps
to spine … Compute

6 * 40-port InfiniBand HDR unmanaged switch QM8790 Storage

The overall topology adopts fat tree with 3:2 blocking ratio.
Interconnect
Spine Spine 01 Spine 02 …
16 * HDR200 uplinks and Summary
24 * HDR200 (48 * HDR100) downlinks

2*8 HDR 100Gbps

(Or 200Gbps)
to nodes
Leaf Leaf 01 Leaf 02 Leaf 03 Leaf 04 Leaf 05 Leaf 06

48 48 48 48 48 48

…
12 * X6000 chassis (48 * XH321 V5 compute nodes) 12 * X6000 chassis (48 * XH321 V5 compute nodes)

Group 1 Group 6
Compute

16
Components quick spec
V5 Typical configuration

X6000 V5 (Optional Liquid cooling) Workflow

2 * CPU
12 * 16GB 2933MHz 2Rank
Compute 1 * 240GB SSD Compute
1288H V5 1 * GE/10GE onboard/NIC
1 * IB EDR/HDR100 card ^ Storage
Server E9000 + CH121 V5
FAT node 2488H V5 24 * DIMM Interconnect
G5500 + G530 V5 2 * CPU
G5500 + G560 V5 16 * DIMM
Acceleration node Summary
2/4/8 * GPU ^
2288H V5 1/2/4 * IB card ^
Management/Login 2288H V5
OceanStor 100D
Huawei Storage
OceanStor 9000
Storage 1 * MetaData building block : 2*2288H V5 + OceanStor 5310 F V5
Lustre
N * ObjectStorage building block : 2*2288H V5 + OceanStor 5510 V5
Reference Architecture
Lustre file system ^
InfiniBand EDR
Interconnection Cable/Switch(including service)/UFM(Optional) ^
InfiniBand HDR100
Network GE - S5731-H48T4XC
Management network
10GE - CE6881-48S6CQ (Optical) / CE6881-48T6CQ (electrical) / S6730-H48X6C
IPMI/Service network GE - S5731-H48T4XC
Cluster management Bright Cluster manager / xCAT etc
Job scheduling SLURM / PBS Pro
Compiler Intel OneAPI / GNU
Middleware
Math library Intel OneAPI / GNU
Message Passing Interface Intel OneAPI / HPCX / OpenMPI
OS CentOS / Ubuntu / SLES

Preferred product / technology ^ Third party integration

17
Requirement questionnaire
Workflow
General Technical Site survey
Compute

Business model Cluster performance Peak value Infrastructure Storage

what are the customer application in use? With detailed Overall (Tflops) Please provide overall DC space and layout.
version Cluster efficiency How much space are planned for the upcoming project?
Qty of Compute nodes Interconnect
what is the workload characterization (percentage of In terms of racks and unit space in rack.
application usage)? Compute nodes form Does the customer going to procure new cabinet/rack?
For each application, please specify the job size (in terms Compute nodes type What is the overall and per rack power supply ability? Summary
of # of nodes/cores). Compute CPU Including PDU information
For each application, how many instances of job will be Node Preferred(Frequency/Cores/Cache) How does the DC being cooled? And what is the
running simultaniously? Memory capacity/frequency/GB per maximum heat dispension rate per rack?
What is the licensing model of customer application? DIMM Requirement What is the cabling method? Underneath floor or
(unlimited, annual, per node/core based) Local storage above rack (with detailed measurement)?
Do all application run on the same hardware platform? Compute
(Same CPU, RAM, network and etc) Type of compute network and speed General Considerations
Network 1.Device deployment should take many components into consideration, such as space taken, device weight,
HPC Cluster Storage usable capacity (TB) power, cabling and many others.
Existing system Storage 2.The positioning of cable lane should be confirmed first, it is a dependent variable for switch positioning
infrastucture Type of parallel File system and cabling.
What is scale of existing cluster? Including node quantity, system 3.Switches should be positioned in central rack of each role to shorten the overall cabling distance.
Storage performance (GB/s)
4.KVM or control station shold be placed at 19U position from bottom to suit operation.
computing ability, power consumption and etc Performance Peak value (Tflops)
Category Description Remarks
Data center information
please provide detailed configuration Model of GPGPU required/Preferred Data center location:
What is the interconnection being used in the cluster? GPGPU Node □ Ground Level | □ Other level
GPU interconnect If data center is not located in the groud level, please check
Technology (InfiniBand, Omni-Path, Ethernet), topology, (If Needed) with customer is elevator or climber available in the building.
Qty of GPGPU Node
blocking ratio Overall data center space:
Basic informatin □<20㎡ | □ 20~50㎡ | □ 50~100㎡
Qty of GPGPU per Node
What is the storage being used in the cluster? Vendor, □ 100~200㎡ | □ >200㎡
Performance Peak value (Tflops) Available space:
capacity, bandwidth, connection type □<20㎡ | □ 20~50㎡ | □ 50~100㎡
What is the cluster management software and job Qty of Fat Node □ 100~200㎡ | □ >200㎡

scheduler being used in the cluster? With detailed version Fat Node type Data center layout diagram □Y | □N
Fat Node UPS for data center □Y | □N
What is the Operating system being used in the cluster? (If Needed) Capacity/Frequency of Memory per UPS suports AC output □Y | □N

With detailed version Fat Node UPS

UPS voltage： _____________V
UPS provide -48V DC □Y | □N
Capacity/Type/Speed of HDD per Fat UPS maximum power supply： _____________KVA
Node Backup power fuel cell： □Y | □N

18
Content

1 HPC introduction

2 HPC solution decomposition

3 Success case

19
 3.1PB 320
Storage GPU

2.6 33280 1040

PFlops Cores nodes

Green 1.2P
500
Huawei Helps Compute Canada Rank#45
HPL

Build Supercomputing Platform to

Propel Science Research
2.6 PFlops compute scale
CPU & GPU acceleration hybrid HPC cluster

20
Huawei Helps PSNC Build the World's TOP 80 Supercomputing Center
The PUE was decreased to 1.05, the energy-consumption decreased 90%, and the
energy cost was saved by 38%.

21
Swiss Federal Institute of Technology
(EPFL), Lausanne Bolsters Scientific
Research with Huawei HPC Solution
Cluster compute performance up by 70%
Compute efficiency improved
22 to 84.4%
 Technical University of Denmark
Accelerates Material Discovery to Application
Uses Intel OPA 100G Fabric interconnection
Cluster computing performance up by 3x

23
 Max-Planck Institute for the Science of Light
Better benchmark performance and higher quality
Multiple deployment since 2016

24
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright©2018 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
HCSP-Presales Computing
Huawei Server Management Software FusionDirector

Security Level:
 Contents

1 Operation &Maintenance Challenge

2 Huawei Computing Server O&M Solution

2 Huawei Confidential
DC Server O&M Evolution

Intelligent

• Automatic processing based on

Automated system behavior
✓ Massive data collection
✓ Massive experience collection
• Challenges
Tool-based • Transition from using tools with
✓ Data accuracy and formatted
manual intervention to unattended
collection of experience
maintenance
✓ Feature extraction based on
Human ✓ Streamlined process and machine learning
• Partially tool-based operations standardization
labor &
• Challenges ✓ Requires support at architecture level
scripting • Challenges
✓ Success rate
✓ Stability ✓ Success rate
• Standalone script
• Batch-operation ✓ Performance ✓ Architecture support capability
script ✓ Standardization
• Manual configuration ✓ Fragmented O&M tasks

Huawei develops new features for automated, intelligent hardware O&M.

3 Huawei Confidential
Challenge 1: Incomplete, Inefficient, and Reactive Fault Management

Incomplete coverage Difficult to locate

New media and component types as well as product complexity Personnel

disparity
are increasing rapidly, while only about 70% of server hardware Slow
response
alarms (out-of-band) can be managed. Difficult
Difficult locating
monitoring

O&M Various
and
complex
Varied devices
vendors Spare
parts
30% are in-band
alarms, not covered
(SSD, hard drive,
DDR, NIC)
With systems interlinked, a single fault will cause chain reaction,
70% are out-of-band putting troubleshooting personnel in a turmoil.
alarms, covered
(CPU core, DDR, SSD, fan, hard
drive, power, sensor)
Large amounts of alarms generated at the same time, making locating
and handling even more difficult.

Difficult to manage multiple data centers in a unified manner.

Source: Huawei data center survey
4 Huawei Confidential
Challenge 2: O&M Is Costly and Difficult
Long deployment High power Low efficiency Difficult edge
cycle consumption management

Network scale:
45 person- 24/7 manual 10,000+ devices
months 50% maintenance
Infrastructure OPEX for electricity
planning & design expenditure

￥
12 9 million
person- Annual electricity
fee for a 1000 kW
months data center
Edge locations: extreme cold,
Deployment seaside, roadside, mountains
acceptance
• Overlapped subsystems, • Traditional data centers • Siloed data center with • Difficult and costly to
complicated engineering have a PUE of about 1.5 separated management and manually deploy,
interfaces, and long • Difficult to access mains reactive maintenance maintain, and patrol
construction cycle supply • Equipment room depends on massive devices
• Impossible to simply replicate • High electricity fees 24/7 manual maintenance • Frequent updates and
customization, hardware iterations of edge services
installation, and software • Complex deployment
commissioning operations. environments

The preceding data is based on the scale of over 2000 servers in Huawei data center equipment rooms. Source: Huawei data center survey

5 Huawei Confidential
 Contents

1 Operation &Maintenance Challenge

2 Huawei Computing Server O&M Solution

6 Huawei Confidential
Huawei Intelligent Computing Server O&M Solution
Software Type Positioning Feature

O&M brain:
Intelligent fault Intelligent asset Intelligent and
Centralized Centralized server O&M in prediction management automatic deployment
management batches, reduced O&M OPEX, and
improved overall stability of Intelligent version Intelligent energy
FusionDirector
servers in data centers management efficiency management

Baseboard management system: Environment Integration

Out-of-band Independently manages the running
monitoring
Fault diagnosis
interface
environment and hardware status of
management servers, visualizes server
Firmware Configuration Energy saving
iBMC management, and improves server
management management management
ease-of-use and reliability.

Assistant management agent:

In-band Monitors server hardware from OS, Slow disk Performance
Memory isolation
reduces unexpected OS detection monitoring
management breakdown, and implements
iBMA comprehensive hardware Hardware
management. monitoring on OS

7 Huawei Confidential
Management Software Networking

Server 1 Server 2 Server N

OS iBMA OS iBMA OS iBMA

In-band hardware management In-band hardware management In-band hardware management
...
Redfish Redfish Redfish
iBMC iBMC iBMC
Onboard out-of-band management Onboard out-of-band management Onboard out-of-band management

Redfish/
SNMP
iBMA: installed after OS installation
FusionDirector iBMC: shipped onboard with server hardware
Centralized management FusionDirector: installed on a standalone server
or virtual machine (VM)

Web/Portal Redfish/SNMP

O&M personnel Upper-layer OSS

8 Huawei Confidential
FusionDirector: Full-Lifecycle Intelligent O&M System
Intelligent discovery
Intelligent maintenance
Server model, configuration, inventorying,
Fault warning, diagnosis, and locating and retirement
7–30 days earlier prediction of risks Automatic inventorying in seconds,
with 100% accuracy

Intelligent upgrade
Automatic version matching,
unattended upgrade
O&M time for 100 servers
reduced from 1 week to hours
Intelligent deployment
Pipeline-style deployment for
Intelligent energy saving
over 10x efficiency boost Tiered energy saving at single-node, cabinet,
and data center levels
Energy efficiency up by 15%

5 Intelligent Management Features Reduce OPEX by 15%

9 Huawei Confidential
Intelligent Asset Management: Leading U Position Anti-Interference
Technology Enables Network-Wide Inventorying in Seconds
Extensive management is falling short Rack mounting
Automatic inventorying Component asset
recommendation
Scale < 100 cabinets Scale > 1000 cabinets Location tracking Asset report

FusionDirector
management
Cabinet/U position/SN
system
information
Device/Component
information Asset
management box
• Time-consuming and labor-intensive, difficult to ensure U position sensor
information accuracy, severe asset loss, difficult to U position
support operation analysis anti-
interference
U position sensor
technology,
Required person-days for inventorying 1000 cabinets ensuring
150 information
100 accuracy
88
100 75
Electronic label
50 U position auto-identification:
0 The leading U position anti-interference technology ensures the accuracy of information
Huawei Bank customer Internet DC collection. Location and device information is managed and displayed in a unified manner
深圳EDC 工行EDC 国富光启IDC
based on the unique information SN.
10 Huawei Confidential
Intelligent Deployment for 10x Higher Rollout Efficiency
Pipeline-based online deployment, automatic planning by location or SN, and serialized process from IP
address allocation to OS deployment and acceptance

15-step device deployment process 11-step automated deployment

Customer case:
Hardware Server Software Acceptance Deployed more than 100,000
commissioning commissioning commissioning servers. Fully automated
Engineering
functions from IP address
Asset acceptance
planning BMC IP configuration OS installation allocation to OS deployment
and device acceptance
Goods improved rollout efficiency by
demand/transfer O&M system more than 10x.
Asset check interconnection Acceptance report
Rollout efficiency
Asset check (server/person-day)
Configuration & check Software & configuration 150
verification
Installation &
power-on Unified password
change

Onsite acceptance

12

Asset stock-in
Manual
人工 FusionDirector

Manual Performed by
operation FusionDirector
11 Huawei Confidential
Intelligent Energy Efficiency Management: Improves Deployment
Density by 20%+ and Saves Energy by 10%
Equipment room status quo
High maintenance costs Low utilization of resources

Cabinet leasing Electricity fees of self- Power supply

Cabinet space
fees account for built equipment rooms usage of cabinets
usage 40–50%
80%+ account for 50%+ less than 50%

Three-level energy efficiency management architecture

Data center Cabinet-level

energy energy Single-node DEMT
consumption consumption
management management
Identification of zombie servers Dynamic power consumption adjustment: Dynamic frequency adjustment:
By analyzing continuously low-load servers, the Dynamically adjusts the server power capping Automatically detects service loads and dynamically
solution supports service consolidation and saves 6% value based on the service trend to ensure reliable adjusts CPU and memory frequencies to achieve
server resources. power supply. optimal service performance and power consumption.

3D display of equipment rooms: Cabinet-level risk display: Intelligent speed adjustment:

• The power supply status of the equipment room is Power supply, temperature, and device status risks Uses the PID algorithm to dynamically adjust fans
visualized, helping identify power supply risks. are globally visualized. based on 50+ parameters such as service load,
• Analyzes high temperature environments and ambient temperature, and component temperature to
identifies environment hot spots. achieve the lowest power.

12 Huawei Confidential
Intelligent DEMT Improves Server Energy Efficiency by 15%
The Dynamic Energy Management Technology (DEMT) is a set of technologies that intelligently adjust the power consumption of each
component in real time, according to multiple input parameters such as load and ambient temperature.

Proportional-integral-differential (PID)
Adjusting fan speed based Digital thermal sensor (DTS)
on processor temperature

Key components Fan modules

Frequency modulation & Intelligent speed
hibernation technologies adjustment technology

Adjusting fan speed

based on processor load

Intelligent Energy-saving Efficient

Functions are automatically Core components run with the lowest power Service performance is not compromised.
triggered after DEMT is enabled. consumption, improving the server energy
efficiency by 15%*. *Data source: Huawei Server Performance Tuning Lab
13 Huawei Confidential
 Intelligent Fault Management: Transforming from Reactive to
Proactive O&M
Diagnosis

FDM intelligent fault

management engine, Provides online and offline
93% diagnosis diagnosis tests for CPUs,
accuracy for memory, and hard drives,
CATEER faults and automatic software and
hardware fault demarcation
Fully out-of-band
monitoring of
hardware faults
Monitoring Prediction
Among the server hardware
faults in data centers, hard
Use AI algorithms to drive faults account for 48%+
predict risky hard drives
7–30 days in advance

Isolates faulty DIMMs and

Isolation hard drives to keep
Isolates DIMM and services always-on
PCIe faults, minimizing
14 Huawei Confidential service downtime
Intelligent Fault Diagnosis with 93% Accuracy
Complete component monitoring Expert diagnosis and prewarning libraries
1. Monitors the health status of all hardware components and 1. Automatic fault data collection by BMC
sensors in an out-of-band manner 2. Huawei's over 20 years of x86 server fault data accumulation
2. Supports both in-band and out-of-band data links for fault 3. Fault diagnosis by Huawei-developed Hi1710 intelligent chip
data collection, no matter whether the OS is normal or not

Information collection diagram Fault locating diagram

CPU
System normal In-band fault
In-band detection data collection
Raw data
Info
receiving
iBMC Out-of-band
Fault occurs
fault collection
PCIe Others System down

Hi171X
Out-of-band
detection data Parsed data
Parsing Diagnosis
Expert
library
Memory
Output
Storage Prewarning
Mainboard, board power Expert
supply, links, PSUs, and library
fan modules

15 Huawei Confidential
Intelligent Fault Prediction: Predicts Risky Hard Drives 7–30 Days in
Advance
Continuous update of Continuous inference
massive samples optimization
FusionDirector
Real-time collection of drive
Model Inference
training
Real-time SMART information for
model
model analysis on the cloud
Inference
Prediction model trained on Drive status result
data
millions of drives

Prediction precision ≥
Self-evolving Accurate 90%

Prediction ≥ 90%
HDD

Lower multi-drive
HDD
Reliable
simultaneous failure rate
Replace hard drive

iBMC

HDD
Real-time collection 90% less manpower for
of hard drive Efficient fault locating
information online
HDD

16 Huawei Confidential
Intelligent Upgrade with 20x Higher Efficiency
• Automatically downloads and detects the firmware and drivers of servers of different models and manages firmware versions comprehensively
• Develops upgrade policy in a one-off manner, no need for the O&M engineer to prepare the upgrade solution and manually upload or upgrade firmware
• Enhanced batch upgrade capability reduces upgrade time

Example: Upgrade 100 servers

Traditional mode: Servers
R&D Version Support-e Manual/Tool-
engineer release
Manual download based upgrade

Huawei GTS Batch upgrade

engineer 20 servers
Software download, solution formulation, and upgrade plan 3 days

Huawei FusionDirector: One-off upgrade policy formulation Minutes Servers

Huawei Online Upgrade FusionDirector

R&D Version Platform (HOUP) Automatic
engineer release Automatic
upgrade
API download
Batch upgrade
100 servers

17 Huawei Confidential
Intelligent Care: Connect to Huawei Service Center for Proactive
Customer Care
Customer Huawei GTAC

FusionDirector
Internet

E-mail server
eService Huawei technical
eService front-end Customer mailbox support center
Huawei servers Back-end

Customer engineer (supervision) Dedicated VPN channel for remote processing

Dispatch work orders

Huawei frontline engineers Huawei service engineers

Huawei's Call Home function reduces information transfer costs and improves customer experience.
✓ Transfer alarms to Huawei ✓ Handle problems quickly
✓ Create service orders automatically ✓ Customer authorization and supervision
18 Huawei Confidential
Edge Device Management: Unified Management of over 10k Edge
Devices and Central Devices
Atlas management network Features
Center
FusionDirector • Monitoring and management of edge devices
Comprehensive • Hardware information, system status, and
application information

Network • Fine-grained management of user rights and

Private line/Public customized role rights
network • Rights- and domain-based management for 1000
Secure groups
• TLS 1.2 encrypted transmission
• Edge + center two-way authentication for
management

Atlas 500 Atlas 500 Atlas 500

Edge nodes • Automatic device management
Node 1 ... Node N • Batch configuration import and export
Simple (configuration restoration after spare parts
replacement)
Maximum management capability of 10k • Automatic firmware upgrade
edge devices + central devices

19 Huawei Confidential
Edge Service Management: Center-Edge Collaboration for Unified
Application Provisioning and Management

✓ A large number of devices ✓ Frequent service model iteration ✓ Devices deployed in remote areas

3rd party ModelArts/Service model management platform

Northbound interface for Manage container Upload image files

secondary development applications
1. Distributes central management
Huawei management software

Edge application
containers and images based on the
FusionDirector Image
management repository service type of the edge device,
implementing fine-grained and
targeted distribution of images.
Push edge application Report container
Pull
management commands application
status information
image files 2. Provides web management interfaces
for users and REST management
interfaces for third-party platforms.
App
Cloud-edge collaborative Docker
…
agent software Engine
Atlas 500 App 3. Pre-installs the agent of the edge
device in factory
Linux OS

20 Huawei Confidential
FusionDirector Installation Environment Requirements
✓ FusionDirector deploys microservices based on the microservice container
architecture and deploys "OS+container+microservice" as a large package.
VM Type ISO Format Version
✓ FusionDirector can be embedded in the MM920 management module of an
E9000 server, installation free and plug-and-play. 6.0 update3hl
✓ FusionDirector supports VM installation and multiple hypervisor VM formats in
enterprise scenarios. 6.5a
✓ FusionDirector supports deployment with a minimum VM environment of "2C/8
6.5hl
GHz/500GB, single network port" and supports deployment on a laptop. The VMware vSphere
.ova
recommended typical configuration is "4C/8 GHz/500GB, single network port". ESXi
6.5u1hl

6.5u2
FusionDirector.iso
6.7

Microservice Microservice Windows Server 2012 R2

app app Microsoft Hyper-V vhdx.zip
...
Bin/Libs Bin/Libs Windows Server 2016

Docker Engine CentOS 7.6

Host OS KVM .qcow2 RHEL 7.3
Hardware
RHEL 7.5

21 Huawei Confidential
Device Management with FusionDirector
Management features： ✓ Displays information about 60+ server components;

✓ Supports management of mainstream Huawei servers and peer ✓ Supports periodic server performance monitoring, facilitating
vendors' servers ； performance analytics.

✓ Manages up to 12,000 devices;

Management scope （ Please refer to the specification list ）：

FusionServer：V3 （RH1288、RH2288、RH2288H、5288、RH5885、RH5885H、RH8100）、V5 （1288H、2288、2288H、2298、2488、

2488H、5288、5288X、5885H、8100、1288X、2288X、2288X LBG-1）、V6（2488H V6）、E9000 MM910/MM920/MM921
Atlas : Heterogeneous Servers （G560、G530 V5、G560 V5、G2500）、 intelligent edge （Atlas 500 AI Edge Station、Atlas 200 SOC）、Atlas
500 Edge Server、Atlas 800（Atlas 800 3000、Atlas 800 3010、Atlas 800 9000、Atlas 800 9010）、Atlas 900 AI Cluster（Atlas 900 Compute
Node）

TaiShan: TaiShan 100（2280、5280）、TaiShan 200（1280、2180、2280、2280E、5280、X6000 XA320、2180K、2280K、2480、5290）

Other vendors' models: HPE ProLiant DL380 G10、HPE ProLiant DL380 G9、HPE ProLiant DL380 G8、DELL PowerEdge R740
Operation Operation Operation Minimum
Typical 3-Node
Duration Duration Duration Management
Item Management Management
(Minutes/Se (Minutes/20 (Minutes/100 Specifications
Specifications (Sets) Specifications (Sets)
rver) Servers) Servers) (Sets)
Time taken for
accessing rack 0.5 1 5 2000 5000 12000
servers

22 Huawei Confidential
Northbound Interface (NBI)
FusionDirector provides open NBIs, uses the Redfish protocol,
Upper-layer NMS and supports SSL to implement seamless interconnection with
upper-layer network management systems.
Redfish SNMP

Interface Type Description

Queries server health/online status
Query/Configuration
management interface and basic information (SNs,
Fusion Information query
Director Alarm interface
configuration, RAID, etc.); obtains the
infrastructure list.
DB
Configures server power-on and
Setting/
power-off, IP addresses, virtual media
Configuration
URIs, and device connectivity testing.
Redfish SNMP
OS deployment Deploys Huawei server OSs.

Upgrades Huawei server firmware

Firmware upgrade
iBMC iBMC iBMC such as BMC, BIOS, and PICe.
TaiShan 200 FusionServer Pro Atlas

23 Huawei Confidential
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright © 2019 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.
Huawei Atlas Enables Industry Intelligent
Transformation
Huawei Atlas AI Computing Solution
 Huawei Atlas AI Computing Solution
AI Unlocks Boundless Possibilities

2
An Open AI Computing Platform for Intelligent Transformation of
Industries
Intelligent edge solution Data center solution

Atlas intelligent edge platform Atlas distributed training platform

Cluster management/Model management/Data
Industry SDK/Container engine/Basic service repository
preprocessing

Framework MindSpore/TensorFlow/PyTorch/etc. Common components

AXE toolchain/Security subsystem/Unified O&M system
Chip enablement CANN

Device Edge Cloud

Atlas 800 Atlas 300T
704 TOPS INT8 280 TFLOPS FP16
Atlas 200 Atlas 500 Atlas 300
22 TOPS INT8 22 TOPS INT8 88 TOPS INT8
Atlas 800 Atlas 900
2.24 PFLOPS FP16 256–1024 PFLOPS FP16

3 Huawei Confidential
AI Will Reshape Industries

Transportation Electric Power

Finance Manufacturing
Public Security • Road-vehicle • Intelligent booster Internet
• Smart branch • Intelligent quality
cooperation station
• Financial OCR inspection • Personalized
• Object detection • Contactless • Unmanned patrol
• Industrial robots recommendation
• Vehicle recognition identification • Intelligent PV
• Content analysis
• Autonomous driving

AI readiness empowers industries

Speech recognition Machine vision Decision and inference Natural language processing

4 Huawei Confidential
 Transportation

5 Huawei Confidential
Huawei Atlas Highway Free-Flow Tolling Solution

AI algorithms and applications

Industrial-grade devices + Docker AI SDK
AI inspection exclusive Docker hardware
Image Edge-cloud collaboration agent
ETC and evidence Data dumping active/standby redundancy
structurization
collection components
Lightweight OS

License plate ETC tolling identification

ETC gantry recognition camera applications

system: integrated RSU antenna MLPS Level III

Networking center of
Province networking
smart site center
Ministry of Transport (MOT)
Cross-province service
Province service processing processing and settlement
Atlas 500
Lane controller
Inspection & Credit big data platform

Traffic efficiency Energy conservation Security control and

up by 5x and emission reduction road collaborative
management

6 Huawei Confidential
Atlas Helps Build a Solid Bedrock for Intelligent Highway
AI Big Data Inspection

Service
application
Cloud-edge collaboration

Credit checking Suspicious ID Image-based Toll evasion control

layer system screening vehicle search
Path restoration Inspection evidence

95%
Platform layer
Cloud Big data inspection
platform
layer Storage/ Integration/
Computing Database AI
Cleaning Law enforcement
against violation

Edge
90%
Analysis
computing License
plate Vehicle data and
structurization evidence Others
layer Atlas 500 AI Atlas 800 edge recognition
collection
edge station inference server

Road network
Data incident analysis
collection
50%
layer
Gantry Toll gate Rest area Surveillance
AI evidence collection AI alert deployment AI warning AI collaboration

7 Huawei Confidential
AI Big Data Inspection Process
Analysis and Toll evasion control Toll evasion Inspection capability upgrade
evidence collection recovery
Vehicle evidence collection Inspection and Evidence chain Graylist update in the
alert deployment generation credit system

Image-based vehicle search

Locking the evidence chain to
Pushing suspicious vehicle data to provide basis for background fee
networking centers on road sections deduction
or local networking centers

Suspicious behavior
Exception detection
AI-based recording of model update
in rest areas
Path restoration missing tolling information

8 Huawei Confidential
Atlas Video Cloud Solution Architecture: AI and Cloud-Network Collaboration
for More Efficient Highway Traffic Monitoring and Control

Command Vehicle privacy Automated Road network AI-aided decision- Mobility service
dispatching protection inspection monitoring making

Atlas key benefits:

Video transcoding Video quality analysis Incident detection Road network monitoring

Algorithm Track analysis

Vehicle collision
Frequency analysis Behavior analysis
Personal information Vehicle information
Others
model analysis archiving archiving

Atlas 300 AI accelerator card

Clouds of province
and MOT
Cloud-edge collaboration
Atlas 800 AI server

Private communication network (SDH, MSTP, and MSTP+)

Checkpoint camera Video surveillance ETC gantry Hand-held device

Convergent sensing Cloud-network Innovative Independence and security

AI-powered precise sensing capabilities collaboration application Huawei-developed key system chips
Secure and trusted cloud- In-depth application of
network collaboration multi-dimensional data

9 Huawei Confidential
Video Transcoding: Service Video Stream Transcoding
for Video Sharing
Device Networking center of road section Province cloud platform Cloud platform of MOT

Huawei • 1080p H.264, 3–12 Mbit/s • 32K H.264, 1 Mbit/s • 32K H.264, 1
• GB/T 28181 • RTMP Mbit/s
intelligent HD
• HTTP-FLV/HLS

MOT
Huawei NVR
• Video storage Intelligent video
• Video stream pulling
• Camera management
gateway Video sharing
service

Video stream
management

Video quality
and control
Transcoding

metadata

Camera
Camera

control
control
NVR
Third-party
camera • Video storage Atlas 300
• Video stream pulling Content delivery
• Camera management Live service
network (CDN)
Client
Atlas 800

Benefits
Hardware video stream decoding
Unit: channel capabilities of a single server
Real-time transcoding HD and low bit-rate

5x ↑
Real-time transcoding for AI hardware
Decoding and re-encoding acceleration and narrowband HD capabilities
1080p H.264/265 real-time transcoding Bit rate down by 20%–30% for the same
video quality
Atlas 800 Vendor

10 Huawei Confidential
 Manufacturing

11 Huawei Confidential
AI Machine Vision for Universal and Accurate AOI

Sample Inspected product Difference is defect.

- =
 The manual inspection is not accurate, which may Higher efficiency Accurate
cause false detection and excessive error tolerance. inspection results
 Knowledge is not quantifiable and experience cannot
be accumulated.
Working principle of traditional AOI:
Finding differences

Working principle of AI-based AOI:

First pass rate: 99%+ AI AOI
Finding similarities

99%
100% coverage of
Inspected Defect
quality inspection
product database

Algorithm model

Parameter Computing Model

Data set tuning verification

 Large data volumes and multiple model iterations

improve inspection accuracy.
 Knowledge is quantifiable and expertise can be reused. Note: Automated Optical Inspection (AOI)

12 Huawei Confidential
AI Machine Vision Quality Inspection Handles Complex and
Diversified Manufacturing Scenarios

PV solar panel
modules

Automobiles and Textile and printing

components

Computer, communication, Medicine

and consumer electronic
products

For complex and unquantifiable patterns, the AI-empowered defect recognition is more
efficient, cost-effective, and precise than traditional machine vision.
*Defect types: material missing; improper material use; edge defect; surface defect or dirt; defect on
appearance, color, pattern/texture, and character, etc.

13 Huawei Confidential
Huawei Atlas AI Quality Inspection Solution

Production scenario O&M scenario

Cloud-edge collaboration
Factory Manufacturing equipment
Manufacturing
Execution System
Public/Private cloud
(MES)
Quality inspection data Edge data return
Deep learning pushed to devices
Keyboard/Mouse

Human-machine Industrial Model

interface Camera generation
computer

Light source Lens

Edge node Model training Model design

Automatic delivery of
models and management Automatic
Pipeline Edge inference for
Touchscreen PLC applications Offline inference & analysis Data prediction
workbench quality inspection optimization

Cloud-edge collaboration for an intelligent factory solution

Continuous model Cloud-edge Ultimate

iteration and update consistency performance

14 Huawei Confidential
Electronics: AI Empowers Intelligent Quality Inspection for Huawei
Southern Factory Intelligent Computing Product Line
Solution deployment Intelligent detection of labels and nameplates

Ubuntu industrial
computer
Atlas 300

Windows industrial
computer

Camera PLC
Electromechanical
control

Low false High detection Fully automatic

positive rate rate inspection

<1% 99.9% 24h

15 Huawei Confidential
 Semiconductor: Intelligent Analysis of Wafer Defects

Data storage cluster

Defect analysis and test equipment
Wafer defect detection (traditional AOI)
40% Image data Data read/write

Closed-loop
quality control
Time for
measurement/semiconductor Algorithms
production AI inference & Models AI training
Atlas 800 AI Server
(Model: 3010) Atlas 800 (Model: 9000)

Wafer production line

Shatter
defect analysis bottlenecks Industry computing density 1388
of manufacturing industry 2x images/s

16 Huawei Confidential
 Finance

17 Huawei Confidential
AI Enables Smart Branches and Fully Taps into Video Data
Employee management Office management Self-service Precision marketing
ID recognition
• VIP • Status and dress code • Employee attendance • Intelligent access control • Customer attribute
check • ID recognition access • Person-certificate identification
identification control verification • Accurate customer profiling
• Potential
customer
identification Branch
functional areas

Human body recognition

Office area
Business
hall
• Tack
analysis
• Attribute Lobby

analysis Entrance

Behavior analysis

Action Potential customer

analysis Operation compliance Security management VIP identification
identification
• Customer dynamic track,
• Stranger entry • Abnormal behavior • Automatic notification for
density, and traffic
• Improper storage of articles • Blacklist check and alert VIP customers
• Quick service adjustment

18 Huawei Confidential
Networking Design and Solution Competitiveness
Compliance check Traffic analysis
Head office DC
Application Training
Storage server FusionDirector platform
system

Model file

Live video display Customer profile

Service Model push
information Status monitoring

Real-time
Branch 1 result
RTSP
Display terminal Atlas 500 Video stream Camera
Hardware reuse Software/hardware
... decoupling
Real-time
result Quick deployment, multi-protocol Software/hardware open and
Branch N support, hardware reuse, low decoupled, multiple algorithms
RTSP power consumption, small running at the same time, and a
Display terminal Atlas 500 Video stream Camera footprint, and low O&M costs unified and efficient hardware base

Cloud-edge Leading algorithms

collaboration
Branch A

Branch B

Branch N

... Data processing is moved forward to Industry's most powerful

reduce the pressure of the central algorithms for customized
equipment room. Low-traffic solutions that best fit
transmission, centralized management customers' needs
and control on the cloud, and
19 Huawei Confidential integrated training and inference
OCR Solution Powered by Huawei Atlas Improves
Service Efficiency
HTTP POST JSON
STR: OCR based on deep learning

Application
RESTful API
interface Handwritten digit recognition Text recognition in natural scenarios

Business Driving Contracts Financial Document type identification Text recognition in complex background
Recognition Bank card VAT invoice
content license license and forms reports
Natural template identification tool Table tool

Core Text direction Character Structured

OCR
algorithms recognition location analysis

Platform TensorFlow
Resource & energy Equipment room
consumption space
Physical Huawei Atlas 800 AI server
hardware Models: 3000/3010
30% 60%

Private High
Fast recognition Format-agnostic Intelligent High integration
deployment recognition rate

Private deployment, Over 98% recognition rate Recognition speed of a Supports scanned images, Automatic, intelligent, and Outputs in Word/Excel,
ensuring user data single A4 image: < 1s images on mobile devices, non-stop around the clock double-layer PDF, and
security JPG, BMP, TIF, PNG and standard JSON and XML
PDF formats; images formats, facilitating
rotated at any angle secondary development
and integration

20 Huawei Confidential
OCR Based on Deep Learning Delivers Better Accuracy
and Performance
Traditional OCR Deep learning-based OCR
(based on CPU computing power) (based on NPU computing power)

Missing information Precise Chinese character recognition rate > 99.3%

Digit recognition rate > 99.8%
A large part of the algorithm is analyzing binary images, and
recognition based on the binary images will cause a large English letter recognition rate > 99.5%
Whole-line recognition rate for handwriting > 95.2%
amount of information loss.

Low recognition rate Fast Average response time < 600 ms

The result can be obtained only after six key steps. The
accumulated error is large. Any incomplete step leads to
error accumulation. The recognition rate of driving licenses
and vehicle licenses is only 70%. Accurately identify images in various
Strong anti-
conditions, such as shadows, wrinkles,
interference reflections, and rich colors.
Poor anti-interference
Complex samples in natural scenarios cannot be processed,
and text in fuzzy and wrinkle scenarios cannot be identified.

Blurry and shadow Wrinkle Multiple bills and

multiple angles
Texts and images must be well illuminated, text must be neat,
21 Huawei Confidential
and text directions must be consistent.
 Electric Power

22 Huawei Confidential
Power Grid: Industry's First Intelligent Unattended Inspection Solution,
with 5x Efficiency
Intelligent analysis at the edge with ultra-
low consumption

Front-end analysis and return 15 W 16 TOPS

Power consumption Computing power
alarms Power consumption reduced
Intelligent camera Atlas 300/800 by 80%
AI AI model training
Model optimization and 3W 2.x TOPS Meeting edge computing
remote upgrade Surveillance and management
platform requirements
Average system Real-time High computing
consumption analysis power

Atlas 200 1W Once per minute 16 TOPS

Manual AI
inspection inspection

Efficiency 5x+ up

System cost 30% down

23 Huawei Confidential
Seamlessly Embed AI Without Changing the Existing Device
Networking, Improving O&M Efficiency of Substations
Traditional monitoring system Atlas intelligent substation monitoring solution

Cloud-based AI Cloud-based AI
Service inference
application training
Service application Atlas 800 Atlas 500 Pro
Monitoring center Monitoring center master station
master station
AI embedded Electric power communication network
Electric power
communication network
seamlessly
AI edge station Edge AI
NVR
Atlas 500 NVR inference
Atlas 500 Pro
LAN/Wireless network

. LAN/Wireless network

. ...
110 kV/35 kV
substation . 500 kV/220 kV
substation 110 kV/35 kV
substation
500 kV/220 kV
substation
Comprehensive monitoring Real-time and intelligent Process visualization and
and intelligent warning edge-based analysis intelligentization
➢ Perimeter warning for ➢ Abnormal device status ➢ Safety helmet detection
restricted areas ➢ Damaged device ➢ Safety dressing detection
➢ Abnormal vehicle appearance ➢ Safety tool usage check
➢ Unsafe acts ➢ Meter recognition ➢ Operation standardization
➢ Smoke, fire, ice and snow ➢ Warning sign identification
detection
24 Huawei Confidential
Unmanned Intelligent Surveillance of Substations

Pointer meter Linked video Operation monitoring

surveillance system center
Digital
Resource
dashboard management
Data interface
AI computing
devices
Indicator Anomaly Real-time
Real-time data detection alarms

Switch Visible light+infrared transmission

position inspection robot Image Image
Video Edge retrieval storage
and image computing Scheduled data
Plate location transmission
data collection Statistics Intelligent
analysis reports
Switch cabinet Video analysis
doors Fixed visible-light camera Correlated
Value mining
analysis

Scheduled Fixed-point Fixed-route Target

Data warehouse
inspection photographing navigation identification
Device Video
Environmental
monitoring surveillance
Identification Infrared Infrared monitoring data
Gas detection Alarm beacon data data
thermometer positioning
label

25 Huawei Confidential
Application of AI in Power Distribution Room

Distribution Power distribution

transformer cabinet (PDC)
Box
camera

Dome
camera

Security & access Equipment Operational standards

control status
• Intrusion detection • Switch status • ID recognition
• Perimeter pre-warning identification • Safety helmet identification
AI edge station • Access control monitoring • Meter reading • Work area perimeter monitoring
• Remote operation
Atlas 500 monitoring

Open: Collaborative:
Integrated algorithm Edge-cloud collaboration
warehouse and support for and remote algorithm
third-party algorithms update

Device status Smart Easy-to-install:

monitoring security Integrated: Industrial-grade, small
Master Intelligent Recording
Compute, storage, size, suitable for power distribution
inspection Big data/AI management
room installation
station communication, and gateway

26 Huawei Confidential
 Public Security

27 Huawei Confidential
Service Scenario: Public Security Converged Analysis
(Info Technology and Video Investigation)
Object search by image: vehicle, pedestrian, and
bicycle/tricycle
Extract feature values for comparison based on the vehicle
Checkpoint e-police Atlas 300 appearance and personnel behavior attributes.
Intelligent recognition server
Improves efficiency and
reduces costs.

Analysis and judgment: Perform big data analysis after

secondary identification.

Video surveillance Big data comparison system

Compares the extracted feature values.

Implement vehicle and human detection deployment

based on driving tracks in videos.
Feature value match

Feature value
database

Benefits

• Unified video application interfaces and modular investigation strategy models support all police applications
• Application-level encapsulation of video interfaces, facilitating upper-layer service development
• Each line of police service and video applications mutually drive each other, promoting continuous development of police video applications
• Atlas offers ultra-high performance to enable fastest person and vehicle search
• One device integrates multiple core applications such as parsing, retrieval, reverse image search, and alert deployment
28 Huawei Confidential
Service Scenario: Traffic Converged Analysis
Crossroads Control key vehicles, such as vehicles carrying dangerous goods,
Atlas 300 dump trucks, and tricycles
Intelligent recognition server
Improves efficiency and
reduces costs.

Highways Detect traffic violation, such as fake license plate, not wearing
Big data comparison system a seat belt, and calling while driving
Compares license plates and vehicle models.

License plate and vehicle model

information storage
Real-time vehicle image database Vehicle dashboard for multi-dimensional data
Main roads
aggregation, analysis, and display
Vehicle registration library
in traffic management
bureaus

Benefits

• Traffic violation detection: security risks such as calling while driving, not wearing seat belts, and illegal vehicles such as vehicles with fake or identical license plates
• Intelligent pre-warning: Associate with service databases of key high-risk personnel and vehicle management offices, intelligently track key vehicles in real time, and
automatically push fake license plates. For high-risk vehicles such as vehicles, obtain the information about the vehicles that are involved in cross-location crime
committing in a timely manner.
• Risk monitoring: Monitor key vehicles such as dump trucks, dangerous goods-carrying vehicles, and tricycles in real time, and warn against risks such as large bus/taxi
gathering, road congestion, and winding vehicles.
29 Huawei Confidential
Service Scenario: Smart Campus & Community
Gates, perimeter, and campus
passages
Smart campus integrated
Mobile phone apps
Key areas and parking lots, etc. application platform

Human / vehicle collection

devices

Campus building entrances

Atlas 500
and gates, etc.

Identity check and turnstiles

Open space in campus

Campus ID recognition Campus crowd analysis Campus vehicle
management

HD video collection devices

Benefits

• Deploy vehicle collection devices at key locations such as campus or community gates, perimeters, and passages to implement alert
deployment and vehicle analysis and management.
• Deploy identity check and turnstiles at campus gates and key buildings to implement ID registration or access control.
• Deploy HD IPCs in open squares to perform intelligent analysis in scenarios such as crowd gathering, over-dense crowds, and disorder.
• Integrate ID recognition, crowd analysis, and vehicle management into the comprehensive application platform (each subsystem can
run independently) to implement intelligent campus and community management.

30 Huawei Confidential
Public: AI Enables Green Smart Campus

As-is To-be Efficient: Recognition and access within 1.5

seconds, 5x efficiency

Forgetting to bring a card ID recognition-based access

Swiping cards for others On-duty attendance Eco-friendly: Overall energy consumption
reduced by 10%, operation cost reduced by 20%

Secure: Incident response time reduced from

Reservation in advance,
Queuing in peak hours person-certificate 7.5 minutes to 2 minutes, handling efficiency
verification improved by 50%

Identifies a wheelchair,
Strollers, wheelchairs
and guides the visitor to an IPC
inconvenient travel
elevator

Access control
device
Crowd traffic prediction Daily crowd traffic statistics
Atlas 500
based on experience Data-driven operations
AI edge station
Smart devices

31 Huawei Confidential
 Other Industries

32 Huawei Confidential
Medical treatment: pilot project of YITU medical new coronavirus
intelligent evaluation system in Rome Hospital, Italy

Intelligent Evaluation System of Novel Coronavirus in YITU Medical Treatment

AI enablement:
Project background: screening of lung slice Report

After the outbreak of the epidemic in China, YITU Medical develops the
Intelligent Evaluation System of New Coronavirus Pneumonia Based on CT
Medical Images (for Medical Institutions) and the Development and
Application of New Coronavirus Pneumonia Prevention Products Based on Huawei AI + Doctor

Speech Recognition, Natural Language Processing and Medical knowledge AI-assisted diagnosis: 10s
Doctor's confirmation: ~ 2 m
graph Technology (for Patients). Automatic report generation: 30s

Pain points:
Doctor

• Imaging is an important basis for diagnosis and treatment of

neo-crown pneumonia. The epidemic has developed rapidly, the Doctor diagnosis: 12 m (min)
Typing input: 2 m (min)

workload of frontline doctors has increased sharply, and medical

personnel are extremely tired.
• Doctors can hardly use the traditional manual analysis method • Video watching speed: 40 pcs/day -> 1000 pcs/day
• Precise quantitative analysis, improving pneumonia diagnosis and treatment capabilities
to accurately analyze the disease condition. However, the
traditional method is time-consuming and inefficient.
YITU + Huawei Atlas 800-3010, 1 to 2 servers for each hospital

33 Huawei Confidential
New Retail: Atlas 200 Implements Intelligent Upgrade of
Singapore DIGI Electronic Scales
Common electronic scale
cannot meet the
AI enablement: requirements of unmanned
Electronic Scale Intelligent Upgrade supermarkets.

Project background:

• DIGI is one of the top 2 electronic scale manufacturers in the world. It

has almost monopolized the supermarket electronic scale market in
Singapore. In addition, Singapore accounts for only about 10% of its
business in Asia Pacific. There is a huge market space in the future. At
about 10 scales per supermarket, Singapore is expected to have at least
5,000+; Demand for AI electronic scales is high.

Pain points:

• Unmanned supermarket price hits the problem seriously and

the peak hour calls the queuing problem;
• In addition, DIGI has a dedicated software and hardware • Powerful computing capability and low power consumption: 8 TFLOPS 9.5W VS
design team, which has software porting and hardware 1.3TFLOPS 15W
design and development capabilities. However, DIGI has just • Improved identification accuracy: N (precision: 80%) → Atlas (precision: 90%)
started piloting AI electronic scales. • The recognition speed is improved from 3.54s to 2.64s, which is 1s faster than that of
competing products and 40% faster than that of competing products.

DIGI+Huawei, site scale: 1000 electronic scales in phase 1

34 Huawei Confidential
Thank you. 把数字世界带入每个人、每个家庭、
每个组织，构建万物互联的智能世界。
Bring digital to every person, home, and
organization for a fully connected,
intelligent world.

Copyright© 2020 Huawei Technologies Co., Ltd.

All Rights Reserved.

The information in this document may contain predictive

statements including, without limitation, statements regarding
the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that
could cause actual results and developments to differ materially
from those expressed or implied in the predictive statements.
Therefore, such information is provided for reference purpose
only and constitutes neither an offer nor an acceptance. Huawei
may change the information at any time without notice.