GRAU DATA Scalable OpenSource Storage CEPH, LIO, OPENARCHIVE

GRAU DATA: More than 20 years experience in storage

OPEN ARCHIVE

2009 2007 1992 2000 2004

Mainframe Tape Libraries Open System Tape Libraries Archive Systeme FMA Software HP OEM FSE & FMA ARCHIVEMANAGER OPENARCHIVE

OPENARCHIVE - Homepage

OPENARCHIVE: 2nd generation of archiving software

Start of development in 2000 100 man-years of development Scalable archive solution from 1 TB up to several PB Common code base for Windows and Linux HSM approach data gets migrated to tapes or disks Fileystem interface (NTFS, POSIX) simplyfies integration Support for all kinds of SCSI backend devices (FC, iSCSI) API only necessary for special purposes > 150 Customers worldwide References: Bundesarchiv, Charite, many DMS vendors

OPENARCHIVE: internal architecture

CLI GUI API
Tape Library

HSMnet Event Manager

HSMnet Server
Management Interface Library Agent

CIFS

Client File System

HSMnet I/F

Partition Manager

NFS

Client File System

HSMnet I/F

Partition Manager

Resource Manager Client File Disks System

Back End Agent(s) Back End Agent(s )

Current limitations of OPENARCHIVE

Number of files and performance is limited by native

filesystems (ext3/ext4, NTFS)

Archiving and recalling of many files in parallel is slow Cluster file systems are not supported Performance is not appropriate for HPC environments Scalability might not be appropriate for cloud environments

High performance SCSI Target: LIO

LIO is the new standard SCSI target in Linux since 2.6.38 LIO is devloped and supported by RisingTide System Completely kernel based SCSI target engine Support for cluster systems (PR, ALUA) Fabric modules for:
iSCSI (TCP/IP) FCoE FC (Qlogic) Infiniband (SRP)

Transparent blocklevel caching for SSD High speed iSCSI initiator (MP for performance and HA) RTSadmin for easy administration

LIO performance comparision (IOPS)

Based on 2 prozesses: 25% read 75% write

iSCSI numbers with standard Debian client. RTS iSCSI Initiator plus kernel patches give much higher results.

Next generation ClusterFS: CEPH

Scalable storage system
1 to 1000s of nodes Gigabytes to exabytes

Reliable
No single points of failure All data is replicated Self-healing

Self-managing
Automatically (re)distributes stored data

Developed by Sage Weil (Dreamhost) based on Ph.D. Theses

Upstream in Linux since 2.6.34 (client), 2.6.37 (RDB)

Ceph design goals

Avoid traditional system designs
Single server bottlenecks, points of failure Symmetric shared disk (SAN)

Avoid manual workload partition

Data sets, usage grow over time Data migration is tedious

Avoid passive storage devices

Storage servers have CPUs; use them

Object storage

Objects
Alphanumeric name Data blob (bytes to gigabytes) Named attributes (foo=bar)

Object pools
Separate flat namespace

Cluster of servers store all objects

RADOS: Reliable autonomic distributed object store

Low-level storage infrastructure

librados, radosgw RBD, Ceph distributed file system

Ceph data placement

Files striped over objects
4 MB objects by default
File

Objects mapped to placement

groups (PGs)
pgid = hash(object) & mask

Objects

PGs mapped to sets of OSDs

crush(cluster, rule, pgid) = [osd2, osd3] ~100 PGs per node Pseudo-random, statistically uniform

PGs

OSDs
(grouped by failure domain)

distribution

Fast O(log n) calculation, no lookups Reliable replicas span failure domains Stable adding/removing OSDs moves few PGs

Ceph storage servers

Ceph storage nodes (OSDs)
cosd object storage daemon btrfs volume of one or more disks
cosd btrfs
Object interface

cosd btrfs

Actively collaborate with peers

Replicate data (n timesadmin can choose) Consistently apply updates Detect node failures Migrate data

It's all about object placement

OSDs acts intelligently - everyone knows where objects are
located
Coordinate writes with replica peers Copy or migrate objects to proper location

OSD map completely specifies data placement

OSD cluster membership and state (up/down etc.) CRUSH function mapping objects PGs OSDs

cosd will peer on startup or map change

Contact other replicas of PGs they store Ensure PG contents are in sync, and stored on the correct nodes

Identical, robust process for any map change

Node failure Cluster expansion/contraction Change in replication level

Why btrfs?
Featureful
Copy on write, snapshots, checksumming, multi-device

Leverage internal transactions, snapshots

OSDs need consistency points for sane recovery

Hooks into copy-on-write infrastructure

Clone data content between files (objects)

ext[34] can also work...

Inefficient snapshots, journaling

Object storage interfaces: librados

Direct, parallel access to entire OSD cluster PaaS, SaaS applications
When objects are more appropriate than files

C, C++, Python, Ruby, Java, PHP bindings

rados_pool_t pool; rados_connect(...); rados_open_pool("mydata", &pool); rados_write(pool, foo, 0, buf1, buflen); rados_read(pool, bar, 0, buf2, buflen); rados_exec(pool, baz, class, method, inbuf, inlen, outbuf, outlen); rados_snap_create(pool, newsnap); rados_set_snap(pool, oldsnap); rados_read(pool, bar, 0, buf2, buflen); /* old! */ rados_close_pool(pool); rados_deinitialize();

Object storage interfaces: radosgw

Proxy: no direct client access to storage nodes HTTP RESTful gateway
S3 and Swift protocols

http

ceph

RDB: RADOS block device

Virtual disk image striped over objects
Reliable shared storage Centralized management VM migration between hosts

Thinly provisioned
Consume disk only when image is written to

Per-image snapshots Layering (WIP)

Copy-on-write overlay over existing 'gold' image Fast creation or migration

RDB: RADOS block device

Native Qemu/KVM (and libvirt) support
$ qemu-img create -f rbd rbd:mypool/myimage 10G $ qemu-system-x86_64 --drive format=rbd,file=rbd:mypool/myimage

Linux kernel driver (2.6.37+)

$ echo 1.2.3.4 name=admin mypool myimage > /sys/bus/rbd/add $ mke2fs -j /dev/rbd0 $ mount /dev/rbd0 /mnt

Simple administration
CLI, librbd
$ rbd create foo --size 20G $ rbd list foo $ rbd snap create --snap=asdf foo $ rbd resize foo --size=40G $ rbd snap create --snap=qwer foo $ rbd snap ls foo 2 asdf 20971520 3 qwer 41943040

Object classes

Start with basic object methods

{read, write, zero} extent; truncate {get, set, remove} attribute delete

Dynamically loadable object classes

Implement new methods based on existing ones e.g. calculate SHA1 hash, rotate image, invert matrix,

etc.

Moves computation to data

Avoid read/modify/write cycle over the network e.g., MDS uses simple key/value methods to update objects

containing directory content

POSIX filesytem
Create file system hierarchy on top of objects Cluster of cmds daemons
No local storage all metadata stored in objects Lots of RAM function has a large, distributed, coherent cache

arbitrating file system access

Dynamic cluster
New daemons can be started up dynamically Automagically load balanced

POSIX example
Client MDS Cluster

fd=open(/foo/bar, O_RDONLY)
Client: requests open from MDS MDS: reads directory /foo from object store MDS: issues capability for file content

read(fd, buf, 1024)

Client: reads data from object store

close(fd)
Client: relinquishes capability to MDS MDS out of I/O path Object locations are well knowncalculated from object name
Object Store

Dynamic subtree partitioning

Root MDS 0 MDS 1 MDS 2 MDS 3 MDS 4

Busy directory fragmented across many MDSs

Scalable
Arbitrarily partition metadata, 10s-100s of nodes

Adaptive
Move work from busy to idle servers Replicate popular metadata on multiple nodes

Workload adaptation

Extreme shifts in workload result in redistribution of metadata across cluster

Metadata initially managed by mds0 is migrated

many directories

same directory

Metadata scaling
Up to 128 MDS nodes, and 250,000 metadata ops/second
I/O rates of potentially many terabytes/second File systems containing many petabytes of data

Recursive accounting
Subtree-based usage accounting
Recursive file, directory, byte counts, mtime
$ ls -alSh | head total 0 drwxr-xr-x 1 root root 9.7T 2011-02-04 15:51 . drwxr-xr-x 1 root root 9.7T 2010-12-16 15:06 .. drwxr-xr-x 1 pomceph pg4194980 9.6T 2011-02-24 08:25 pomceph drwxr-xr-x 1 mcg_test1 pg2419992 23G 2011-02-02 08:57 mcg_test1 drwx--x--- 1 luko adm 19G 2011-01-21 12:17 luko drwx--x--- 1 eest adm 14G 2011-02-04 16:29 eest drwxr-xr-x 1 mcg_test2 pg2419992 3.0G 2011-02-02 09:34 mcg_test2 drwx--x--- 1 fuzyceph adm 1.5G 2011-01-18 10:46 fuzyceph drwxr-xr-x 1 dallasceph pg275 596M 2011-01-14 10:06 dallasceph $ getfattr -d -m ceph. pomceph # file: pomceph ceph.dir.entries="39" ceph.dir.files="37" ceph.dir.rbytes="10550153946827" ceph.dir.rctime="1298565125.590930000" ceph.dir.rentries="2454401" ceph.dir.rfiles="1585288" ceph.dir.rsubdirs="869113" ceph.dir.subdirs="2"

Fine-grained snapshots
Snapshot arbitrary directory subtrees
Volume or subvolume granularity cumbersome at petabyte scale

Simple interface
$ m kdir foo/ .s na p / one # c re a te s n a ps h ot $ ls foo / .s na p on e $ ls foo / b a r/ .s na p _o ne _1099 511 627 776 # pa re nt's s n a p n a m e is m a ng le d $ rm foo / m yfile $ ls -F fo o b a r/ $ ls foo / .s na p/ on e m yfile b a r/ $ rm dir foo/ .s na p / one # re m o ve s n a ps h ot

Efficient storage
Leverages copy-on-write at storage layer (btrfs)

POSIX filesystem client

POSIX; strong consistency semantics

Processes on different hosts interact as if on same host Client maintains consistent data/metadata caches

Linux kernel client

# modprobe ceph # mount -t ceph 10.3.14.95:/ /mnt/ceph # df -h /mnt/ceph Filesystem Size Used Avail Use% Mounted on 10.3.14.95:/ 95T 29T 66T 31% /mnt/ceph

Userspace client
cfuse FUSE-based client libceph library (ceph_open(), etc.) Hadoop, Hypertable client modules (libceph)

Integration of CEPH, LIO, OpenArchive

CEPH implements complete POSIX semantics OpenArchive plugs into VFS layer of Linux Integrated setup of CEPH, LIO and OpenArchive provides:
Scalable cluster filesystem Scalable HSM for cloud and HPC setups Scalable high available SAN storage (iSCSI, FC, IB)

I have a dream...
Archive
Tape Disk Archive

ClusterFS

SAN

OpenArchive CEPH

Filesystem Filesystem Filesystem CEPH

Blocklevel Blocklevel Blocklevel Initiator

CEPH MDS CEPH OSD

LIO Target RDB

LIO Target RDB

Contact

Thomas Uhl Cell: +49 170 7917711 [email protected] [email protected] www.twitter.com/tuhl de.wikipedia.org/wiki/Thomas_Uhl