0% found this document useful (0 votes)
149 views9 pages

5210 SSD Vs HDD Endurance White Paper PDF

Uploaded by

Sonam Rai
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
149 views9 pages

5210 SSD Vs HDD Endurance White Paper PDF

Uploaded by

Sonam Rai
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Comparing SSD and HDD Endurance in the

Age of QLC SSDs

SSDs brought unique benefits into client and enterprise storage — high speed, low power and low latency to drive
emerging (and standard) applications toward new performance thresholds. SSDs also introduced a new concept into the
storage market: storage devices wear as they are written and require a warranted write endurance limit (SSD endurance)
as a known variable versus HDD time-based warranties. Different types and sizes of SSDs offered different wear ratings,
but the idea of storage devices wearing predictably as they were used was new. In the early days of SSD adoption
(around 2007), HDDs didn’t have warranted endurance ratings; today, ratings are much more broadly published.

The market saw initial user trepidation, partly because users didn’t know their I/O patterns and had difficulty estimating
them. As a result, many storage architects deployed SSDs with endurance ratings far greater than their workloads
needed.

In this paper we will discuss how SSDs wear versus HDDs and how they can be compared based on workload/application
needs, as well as:

• Developments in HDD endurance statements


• Converting device endurance ratings into common units
• Two examples of HDDs with wear ratings (workload limit ratings)
• How capacity-focused, enterprise SSD and HDD endurance differs
• How the introduction of new QLC NAND has created an endurance paradigm shift

Understanding how SSDs and HDDs are affected by different application environments will enable you to make purchase
decisions based on your workload needs.

Note: This paper is focused on warranty statements related to endurance. These statements are often found in product
manuals or data sheets. It does not contemplate actions taken when those values are reached. Although current at the
time of publication, values used in this paper are subject to change. Your results may differ from those stated herein. This
paper uses the terms “warranted endurance” and “endurance” interchangeably.

1
Storage Wear –
A Printing Analogy
Rated or warranted endurance wasn’t a unique concept in
the computing ecosystem. Printer cartridges are an SSDs and HDDs incur wear differently.
example of computing devices that wear. The two are
(somewhat) analogous in this way. SSDs wear when written (expressed as
drive writes per day or DWPD).
We have long accepted that printer cartridges can print
different numbers of pages depending on how much print HDDs with a workload limit rating incur wear
is on the page. If we print pages that have a lot of images when read or written and are compared on a
and require a substantial amount of ink, we accept that the DRWPD basis (drive reads or writes per
cartridge will have a shorter useful lifespan versus if we day, expressed as a ratio of the drive’s
print pages with a little text. The less we print (or usable capacity).
effectively wear the cartridge), the longer the cartridge will
last.
If you make drive purchase decisions based
on rated endurance, we show that for many
SSDs and HDDs (with wear limit ratings) are like print read-focused workloads, low endurance
cartridges in this way: they wear based on use. HDDs SSDs meet or exceed the DRWPD of some
wear based on data written or read, while SSDs wear capacity-focused enterprise HDDs.
based on writes.

SSD Wear
The NAND storage used in SSDs is different from the magnetic storage used in HDDs. The magnetic storage on HDDs
supports write in place; if there is already data in the physical location to be written, the existing data can be directly
overwritten with new data.

This is a single-step process. However, when data is present in NAND (even data that has been marked as no longer in
use), NAND must be erased before it can be written (programmed). This two-step process is called a program/erase (P/E)
cycle, and SSD endurance is a function of the number of P/E cycles for which the NAND is rated. (Additional details on
this process may be found in this Micron Brief.)

NAND wears when it is written (this becomes important in the following sections), but only incurs negligible wear when
read.

SSD Capability to Absorb Wear


As SSD adoption expanded, the industry grew accustomed to the idea that storage devices had a rated wear value. SSDs
usually refer to this rating as drive writes per day, or DWPD (or less commonly, as total bytes written, or TBW). These
values are related: DWPD = TBW / [(drive capacity) * (warranty period in days)]. We will use DWPD when discussing SSD
warranted endurance.

Because the idea of storage incurring wear when written was new, many system designers overestimated the amount of
DWPD they needed from SSDs. In the early days of SSDs, 10, 20 or more DWPD was normal. It ensured a safety margin.
It was also due to drive capacities being significantly smaller (most early SSDs held less than 1TB). However, trends
showed DWPD rapidly decreasing as drive capacities increased and the industry gained a better understanding of
workload read/write profiles.

Figure 1 shows how DWPD requirements have decreased over time.

2
The type of NAND flash technology that SSDs
are built on differs in several ways. The most 75% Enterprise SSDs Shipped Worldwide (2017): <= 1DWPD
fundamental difference among NAND types is
the number of bits stored in each cell. The
number of bits (0s or 1s) is controlled by the
number of charge states for which the NAND
is designed.

At one end of the spectrum, single-level cell


(SLC) NAND holds one bit per cell. At the
other end of the spectrum, new quad-level cell
(QLC) NAND supports four bits per cell.

These different types of NAND (SLC, MLC,


TLC and QLC) are also typically rated to
support a different number of P/E cycles. (As
any NAND approaches its rated P/E cycle
count, it becomes increasingly difficult to
determine if the data stored is a 0 or a 1.)
There are some techniques that can extend
the number of P/E cycles supported, but at
Figure 1: SSD Endurance Trends
some point, the NAND is worn to the point (Source: Analyst Consensus, Forward Insights Datacenter, May
where it is retired. The state of wear for the SSD 2018)
is managed internally and is easily monitored by
the OS or storage system, making the lifespan
of a drive very predictable.

Generally speaking, as the number of bits per


cell increases, the number of P/E cycles for
which the NAND is rated decreases. This is
due in part to the additional complexity
introduced by more charge states. The effect
is reduced warranted write endurance of
storage devices using that NAND, along with
reduced cost per bit stored, as illustrated in
Figure 2.

Changing Trends in Figure 2: Cost, Complexity and Write Endurance Trends

SSD Endurance
Requirements
Advances in NAND technology spurred the enterprise SSD market to broaden adoption and at the same time rethink
endurance requirements and re-analyze workloads and the amount of data they write. There was a general realization
that many workloads wrote far less than initially anticipated. This drove broad reconsideration of how much write
endurance was really needed. Given the reduced price per gigabyte, combined with increased per-drive capacity
introduced by advances in NAND storage technology, more workloads could take advantage of SSDs. Workloads that
wrote less and benefited from flash (like a read cache) became prime candidates for denser, lower-warranted endurance
NAND.

This culminated in the introduction of QLC NAND and Micron’s release of the first QLC SSD to the market in 2018.

3
Figure 3: SSD Trends

HDDs Adopt Workload Ratings


HDD capacity has also grown in the same timeframe (using different technologies than SSDs). At the time of this paper’s
publication, leading HDD vendors were offering 14TB 3.5-inch form factor products. As HDD capacity grew, the breadth of
HDD types also grew, with some designed for specific use cases. As their capacities increased, some HDD designs
began to adopt a “workload limit” rating as part of their standard specifications.

Workload Limit Ratings


HDD workload limit ratings are very different from SSD DWPD ratings. Both are warranted values (exact warranty terms
will differ between brands, models and technologies), but while SSD warranties typically state endurance in terms of the
amount of data written, HDD warranties typically state workload limits in terms of bytes written and/or read. (The reasons
for HDDs adopting specific workload limit ratings are beyond the scope of this paper.)

This means that different I/O types wear SSDs and HDDs differently, as shown below.

Drive Type Reading Data Writing Data Varies w/Pattern


SSD Minimal/no wear Causes wear Yes
HDD Causes wear Causes wear No

Table 1: Drive Wear Incurred by I/O Type

4
Comparing Endurance: SSD TBW and HDD Workload Limit
In this section we compare the warranted endurance of two typical, enterprise-class HDDs designed for high-capacity
storage.

Each example is an 8TB enterprise-grade HDD, one designed for general purpose storage and the other as a bulk
storage device (active archive). These HDDs are only examples; exact workload limits may vary by manufacturer, model,
capacity, generation or many other factors. Some models don’t have a specified workload limit; therefore, your results
may vary.

HDD Workload Limits


To illustrate HDD endurance, we’ll examine two enterprise-class 7200 RPM drives. Note that other HDD types may have
different endurance ratings. Table 2 shows their relative values and data sheet workload limit ratings. (Workload limit
ratings are typically expressed in TB/year.) Since HDD workload limit ratings include both read and write I/Os, we can
express a workload limit in SSD-familiar terms, or DRWPD. The DRWPD value in the table below was derived from each
drive’s data sheet workload rating using simple math.

HDD Workload Limit Rating DRWPD HDD Capacity Spindle RPM Warranty
(TB Read or Written
per Year)
Example 1 550TB 0.19 8TB 7200 5 Years
Example 2 180TB 0.06 8TB 7200 3 Years

Table 2: Workload Limit Ratings by Example HDD

The DRWPD values noted in Table 2 were calculated as follows:

Example 1 HDD: 8TB, 3.5-inch 7200 RPM Example 2 HDD: 8TB, 3.5-inch 7200 RPM
general purpose enterprise storage active archive enterprise storage

Workload limit rating: 550TB per year Workload limit rating: 180TB per year
Warranty duration: 5 years Warranty duration: 3 years
Drive capacity: 8TB Drive capacity: 8TB
Warranted drive reads or writes per Warranted drive reads or writes per
day: (550TB per year/365 days per day: (180TB per year/365 days per
year)/8TB capacity = 0.19 year)/8TB capacity = 0.06

DRWPD has not historically been a frequently used industry term, but in the age of QLC SSDs, it will likely become a
more important point of comparison, as SSDs increasingly compete with HDDs for primary data storage. We create and
use the term here to enable clearer, more direct comparisons with SSD DWPD ratings.

We now can graph HDD workload limits as DRWPD, as shown in Figure 4.

5
HDD DRWPD by Drive, Workload
8TB HDD Arch 8TB Ent. HDD

0.90
0.80
0.70
0.60
DRWPD

0.50
0.40
0.30
0.20
0.10
0.00
100% 4K RND 100% 8K RND 100% 16K RND 50% 128K SEQ / 70% 128K SEQ / 80% 128K SEQ / 90% 128K SEQ / 100% 128 SEQ
50% 4K RND 30% 4K RND 20% 4K RND 10% 4K RND
Data Write Pattern

Figure 4: HDD DRWPD by Write Pattern (workload)

Note that DWPD is constant for these drives. This is expected since rated endurance is also fixed (workload-
independent). The x-axis shows several different I/O mixes, or data write patterns. Figure 4 indicates what percent of all
write I/O traffic is random versus sequential (and common transfer sizes for those accesses). Figure 4 also assumes that
the HDD can generate sufficient I/O to realize these DRWPD values. This may not be the case with all HDDs.

Comparing SSD DWPD and HDD DRWPD


Earlier this paper noted that NAND wears when it is written, not when it is read. This means that NAND-based SSDs also
wear when written, not when read. (Note that read disturb, a phenomenon of SSDs, incurs a slight amount of wear, but it
is negligible.) SSDs incur more wear when I/Os are small and randomly placed. The opposite is also true — SSDs incur
less wear when the I/Os are large and sequentially placed. (See this Micron Brief for additional details.)

Prior to the introduction of QLC NAND technology, SSDs were rated at a fixed DWPD value. That means that their DWPD
ratings did not change with applied workload. SSD DWPD was typically specified with a small (4K), 100% random write
workload.

The difference between SSD wear applied by different write patterns and SSD DWPD ratings should be understood: while
different write patterns may cause more wear, prior to the introduction of QLC NAND, SSD-rated DWPD did not change.

When QLC technology was first shipped into the enterprise SSD market (the Micron 5210 ION was the first such SSD
shipped), QLC NAND wear characteristics and workload understanding had matured. With that greater understanding
came a new way of expressing QLC-based SSD endurance, and the data sheet for the Micron 5210 ION SSD gave
workload-specific endurance expectations — an industry first that was quickly followed and adopted by other
manufacturers of QLC SSDs.

This means that the rated DWPD on the Micron 5210 ION SSD varies based on the type of write I/O — small and random
I/O means lower DWPD; a heavier mix of large and sequential I/O means higher DWPD.

Figure 5 shows the DWPD of a 7.68TB Micron 5200 ECO SSD (whose DWPD is noted as a constant in the data sheet)
and a 7.68TB Micron 5210 ION SSD (whose DWPD varies with workload), both overlaid on the HDD DRWPD that was
previously noted and discussed in Figure 4.

6
Note that Figure 5 uses “DxPD” to indicate that it shows both DWPD and DRWPD. When DxPD references an HDD, it is
understood to mean DRWPD; when referencing an SSD, it is understood to mean DWPD.

HDD/SSD DxPD by Workload


8TB HDD Arch 5210 7.68 TB 8TB Ent. HDD 5200 ECO 7.68 TB

0.90
0.80 5200 ECO:
0.70 constant DxPD
0.60
DxPD

0.50
5210 ION: workload-
0.40
dependent DxPD
0.30
0.20
0.10
0.00
100% 4K RND 100% 8K RND 100% 16K RND 50% 128K SEQ / 70% 128K SEQ / 80% 128K SEQ / 90% 128K SEQ / 100% 128 SEQ
50% 4K RND 30% 4K RND 20% 4K RND 10% 4K RND
Data Write Pattern
Figure 5: Micron Read-Intensive Enterprise SSD DWPD vs. HDD DRWPD (DxPD) by Write Pattern (workload)

Figure 5 compares only enterprise SATA drives (SSD and HDD) to ensure a fair comparison. The 5210 ION SSD DWPD
ratings trending up and to the right indicate that as the data write pattern changes from small random I/O (at left), to
larger, sequential I/O (right), DWPD increases on the 5210.

The 5200 ECO and HDD DWPD data sheet ratings do not follow this trend. Both are independent of the I/O type with
small random and large sequential, resulting in the same DxPD (appearing as a fixed, horizontal line).

Figure 5 also shows:

• The 5210 and archive-class HDD have similar DxPD if all write traffic is small (4K) and random. When the I/O size
reaches 8KB, the 5210 has higher DxPD (trend continues for all additional write patterns).

• The 5210 and enterprise-class HDD have similar DxPD when write traffic is 90% 128K sequential; the 5210 has
higher DxPD for 100% 128K sequential writes.

• The 5200 ECO DxPD exceeds both HDDs for any write pattern shown.

Using DxPD to Help Find the Right Drive – SSD or HDD


There are three key aspects of this information that will help align specific enterprise workloads to storage devices.

DWPD is Essential in Understanding Endurance in the Age of QLC SSDs


DWPD and DRWPD are ratios of how much of a drive’s capacity can be written (SSD) or read/written (HDD) per day
based on the drive’s usable capacity, so it can be misleading to choose based solely on DWPD values. To best
understand how much endurance one is getting, DxPD should be converted to gigabytes written (or read/written for

7
HDDs) per day. Gigabytes is simply a function of DWPD: If a 7.68TB SSD has a DWPD rating of 0.5 DWPD, it means that
50% of the drive can be written per day (3.84TB).

This is especially relevant when evaluating a QLC SSD since QLC technology packs 33% more bits into every cell and
QLC SSDs are typically only available in higher capacities (which inherently have lower DWPD ratings).

For example, a 960GB TLC SSD with a 1 DWPD rating delivers DWPD similar to a 1.92TB QLC SSD that has a 0.5
DWPD rating for a given workload. While the QLC SSD DWPD specification appears lower, the amount of writes per day
is similar.

DWPD Needs Are Lower for Many High-Growth Workloads


There is a strong trend in high-growth applications showing that they read far more data than they write. Many industry
analyst firms have indicated very high growth rates for read-centric enterprise workloads ranging from artificial intelligence
(AI), machine learning (ML), big data analytics, low-ingest Ceph block/object storage, some NoSQL workloads (profile
caching, read latest, etc.) and deep learning and business intelligence.

These read-centric applications can be a very good fit for SSDs (which incur very little wear when written). However, if
you are replacing drives based on warranted endurance values, you may want to be very selective when using HDDs for
these same applications since many HDDs incur wear when read and written. While some HDDs have no workload limit
rating, many high-capacity enterprise HDDs have this rating, so care must be taken with HDD selection.

Understanding Endurance
Many read-intensive workloads still have write I/O as part of their read/write ratio, even if they’re classified as being
heavily read-intensive (for example, ingesting data into HDFS for subsequent analytics). SSDs can meet and in some
cases, exceed those needs, even though the SSDs are classified as read-intensive drives. Read-intensive simply means
that a greater proportion of the target workload involves reading data, not writing data.

Many read-intensive applications are still deployed on HDDs instead of SSDs, and this may be due to initial endurance
concerns. As we’ve seen, the endurance of the 5200 ECO and the 5210 ION SSDs exceeds that of both archive-class
and enterprise-class HDDs under common use cases (depending on the workload I/O profile).

Conclusion
SSDs and HDDs (with workload limit ratings) incur wear differently. SSDs wear when written, and their wear tolerance is
expressed as DWPD. HDDs are different. HDDs with a workload limit rating incur wear when read or written. In this paper
we express their wear tolerance as drive reads or writes per day (DRWPD).

Their capacities differ as well. Converting DWPD and DRWPD into GB/day values helps normalize the effects of capacity
and makes it easier to compare their wear differences.

We noted earlier that when any drive is deployed in an IT environment — HDD or SSD — the choice depends on several
factors that may vary widely among organizations. Your company may prioritize time, acquisition cost, TCO or other
factor.

8
Understanding your workload by using analysis tools built into many operating systems can also help. These tools can
show you what I/Os are being sent to storage and provide a deeper understanding of how your applications are using
storage and whether the application is more write-intensive or read-intensive.

If you make drive replacement decisions based on rated endurance, we showed that for many workloads, SSD DWPD
ratings are equal to or exceed the DRWPD of some capacity-focused, enterprise HDDs — making SSDs a great fit for
emerging and traditional read-focused enterprise workloads.

micron.com
©2018 Micron Technology, Inc. All rights reserved. All information herein is provided on as “AS IS” basis without warranties of any kind, including any
implied warranties, warranties of merchantability or warranties of fitness for a particular purpose. Micron, the Micron logo, and all other Micron trademarks
are the property of Micron Technology, Inc. All other trademarks are the property of their respective owners. No hardware, software or system can provide
absolute security and protection of data under all conditions. Micron assumes no liability for lost, stolen or corrupted data arising from the use of any
Micron product, including those products that incorporate any of the mentioned security features. Products are warranted only to meet Micron’s
production data sheet specifications. Products, programs and specifications are subject to change without notice. Dates are estimates only. Rev. A 11/18
CCM004-676576390-11207

You might also like