some twisty turny stories of DWPD
Editor:-
this somewhat messy but serious article below - which I've added to over a
period of 4-5 years - started as a simple question about the write limits
of enterprise flash accelerators (PCIe SSDs) used as caches
and virtual memory compared to
SAS / SATA SSDs in
storage arrays.
Then to my surprise it twisted and turned as DWPD discovered roles in other
SSD markets (industrial,
consumer,
cloud) before the
narrative bounced back to generate "no limits"
petabyte AFA good
denizens. But the story of DWPD in the
storage news
archives didn't stop there and grew in the telling. And DWPD didn't stay
rooted in the flash world either and became quoted for other
non volatile memories too
- and in this article there has even been talk of DWPD being a useful way
to talk about
hard drives
too. |
|
what's the state of DWPD?
endurance in industry leading enterprise SSDs (and also DWPD in
consumer SSDs, industrial SSDs and military SSDs too)
by
Zsolt Kerekes,
editor - StorageSearch.com
|
DWPD (Diskful Writes
Per Day) for 5 years - has become an established part of
SSD jargon in the
writings of enterprise SSD makers.
This article began in 2014 when I
thought it curious that I hadn't seen an
endurance
figure expressed in this way (as DWPD) from Fusion-io for its
marketing leading PCIe
SSD accelerators.. Then a
blog
- written in March 2014 by Chris
McCall gave me the answer for the ioDrive2 . It was 4.4.
You may ask... Is that good or bad?
In the number score wars
- there are competing enterprise flash SSD drives which are better. And
some that are worse.
I realized that DWPD numbers were popping up all
over the place. And sometimes the DWPD rating was conspicuous by its absence -
a supporting clue in technically lightweight enterprise SSD press releases
- that the related product launch signified a new model line designation
rather than a fully characterized and shippable SSD.
When I saw how
popular my initial brief roundup of DWPD examples had become - I decided to
update the list from time to time with interesting examples as you can see in
the lists below.
uses and limits of DWPD
DWPD was created to
mitigate the fear that SSD integrators could choose the wrong type of flash SSD
for the anticipated prohect workload. It provided a simple to understand "goodness"
metric by which competing SSD manufacturers were able to signal the endurance
of the SSD which integrated all the combined effects of the internal memories,
interface speeds, controllers and architecture. Before DWPD it was harder to do
this because SSD datasheets were specified in different ways.
The
convenience of DWPD as a way of selecting SSDs for application roles meant it
quickly gained widespread adoption in the enterprise and cloud. Indeed it was
so useful that within 3 years it got picked up as a specification seen in most
of the other markets in which SSDs were used too including:- consumer,
industrial and military drives. That's also why just looking at the single DWPD
rating for an SSD doesn't tell you whether it's a good or bad SSD. You need to
know the context of the application.
no DWPD? - detects vaporware
DWPD
has also been a useful way to recognize products which have been"announced"
far in advance of their actual sampling and characterization.
The
missing DWPD ratings in vendor press releases about new products were a good
clue to products which you were not likely to see for many quarters (if indeed
ever).
dwpd - let's calm it down a bit
Now before
this starts to sound too much like a DWPD fan club I'd like to say that -
like all technical specifications - DWPD has its uses but also has its
limitations too.
DWPD for RAM?
Created for
traditional storage roles it's questionable how useful a DWPD rating is when
it's provided in the context of an
NVDIMM which
is intended for use as a memory (tiered flash emulating RAM). That's because
the R/W and caching characteristics of flash deployed as RAM are installation
and applications dependent. And there haven't been enough examples in the market
to determine what are the useful specification shortcuts for choosing between
competing devices. If there are any I suspect that DWPD won't be top of the list
and even IOPS are
dubious in this context. Having said that - the DWPD ratings which have
appeared in various RAM roled SCM products do still provide an indication
that the designers are trying to allay fears about wear-out.
no-limits
DWPD
Meanwhile - slower down the throughput curve but higher up
the drive capacity - in Q3 2017 we
started seeing
the emergence of SSDs with "unlimited DWPD" - in the shape of a
high capacity SAS SSD platform from
Nimbus which has no
write limits for 5 years.
In part this is the consequence of using a
comparatively slow - 6Gbps - interface and connecting it to enough flash chips
in a 3.5" form factor that you would struggle to get high drive writes per
day even if your tried hard - and the rest is the use of high quality flash and
good controller management. But for the intended roles of such SSDs - which are
high capacity storage drives - the assertion of "unlimited DWPD" shows
that every specification has its uses and limits.
DWPD - a list of
products from the
news archive
(2014 to 2017)
- from Seagate
- the
1200
SSD (12Gb/s, eMLC SAS
SSD) is rated from 10 to 25 DWPD - depends on model - for 5
years
- from Intel -
the
P4800X
(a 3DXpoint / Optane memory PCIe SSD) rated at 30 DWPD for 3 years - which is
18 DWPD for a 5 year comparison basis.
- from Intel -
the
DC
S3700 (1.8" and
2.5" 25nm "older
technology" - my words / "high endurance technology" - Intel's
words - SATA 3 SSD)
is rated at 10 DWPD for 5 years
- from Mangstor
- the MX6300 - a
fast NVME PCIe SSD FHHL gen3 x 8 which uses eMLC and 100% overprovisioning
and has a proprietary internal 100 core controller - is rated at 7
DWPD for 5 years.
- from Virtium
- the "XE models" in its
StorFly (form factors
from 2.5" down to M.2 range of iMLC rugged industrial SSDs) are rated
at
7 DWPD for 5 years.
- from Solidata
- the
Rana
(a military grade, industrial temperature 2.5" SATA 3 MLC SSD) is rated at
5 DWPD for 5 years.
- from Micron
- the
5100
MAX (a 2.5" SATA
SSD using 3D TLC) is rated at 5 DWPD and in March 2017 Micron
claimed
this was one of the fastest SATA enterprise SSDs on the market with steady
state random writes (up to 74,000
IOPS - 4KB).
- from TCS
- the
Triton
2 (a military grade, industrial temperature 2.5" SATA 3 MLC SSD) rated
at 1.4 DWPD for 5 years.
The real answer to - what's
good enough? - when it comes to DWPD - depends on the application.
PS
- before you get too hung up on a particular value of DWPD...
A reader
told me before I published the original version of the above article - that
he was interested in knowing DWPD at different operating temperatures - within
the industrial and military range. (The reason being that in his projects it
would be known in advance - at what temperature the SSDs would spend most of
their operational lives. He didn't have much confidence in the numbers which
various vendors had given him. This doubt was fueled by having being told widely
different ratings for SSDs which had apparently identical controllers and
memory.
I added to that skepticism when I said that I wouldn't be
surprised to see DWPD ratings drift in a more favorable direction even in the
case of the same model SSD - even when nothing had actually changed in the BOM.
My reasoning was that DWPD - like all endurance figures - is the
output of negotiations between engineers and marketers in the same company. With
new products and scant hours of validation - it's natural for engineers to
want to specify a cautious figure - which tends to pull the rating down. But as
time goes on - and more evidence is gathered - and firmware changes get done -
and the manufacturing tolerances in the memory wafer fab becomes more tightly
managed - it may be technically justified to publish a much improved DWPD
rating compared to the original version - even when it looks like all the
components are exactly the same as they were before.
Competitive
pressures from the market play a big a part in the DWPD numbers you see being
published in the enterprise. It's a fashion statement tied in closely to
design wins. If you've got the wrong number - the customer design team is not
going to look at your SSD. DWPD is a slippery relativity thing rather than
a hard and fast technical absolute.
You might think that marketers always want the highest rating which
engineering will allow them - so as to justify a higher ASP - but marketers
may also push to offer good models with a much lower rating so as to position
the product with a lower price for a different application segment. In the
latter case - the lower DWPD rating isn't a guarantee that the product you buy
today will be substantially worse in real life. A low rating can be applied
to a higher rated product to temporarily test the market reaction for a cheaper
(gap filler) product - in a way which protects the brand investment in the
higher rated models. | | |
| ... |
 |
| ... |
metascale advised, life-cycle fitted
the new virtually hardened flash endurance
another plot twist in the continuing elasticity of flash's DWPD story
|
Editor:- April 28, 2016 - There
was a time when you could write a simple list of nand flash memory types
which you'd be likely to encounter in the market and alongside the expected
endurance.
We know how these lists look - because we've seen them so
often.
Often the lists are presented as graphs instead - because they
brighten up a web page or powerpoint.
The list goes something like:-
SLC, MLC, TLC, then for each of those a note saying if it's 2D or 3D, along
with an entry for each nanometer cell geometry.
Then alongside
that list you have the range of likely endurance (P/E cycles).
There's
widespread agreement about what these numbers should be. And if we see the
wrong number in the wrong place we know that this was probably a simple
transcription error.
These hard endurance expectations - when applied
to an SSD content - got a bit of a reality check back in 2011 when - as I
commented in my article the
new SSD uncertainty principle - we started to learn more about some of the
stretching effects of endurance which were being enabled by a mix of technology
tricks which were later called
adaptive R/W
and DSP ECC IP. It became clear that this pioneering approach would
become the standard way that many controllers used.
Recently the SSD
market has become accustomed to compartmentalizing the raw endurance figures
attached to memory chips (on the one hand) and the endurance magnification
effects of adaptive controller techniques (on the other).
But I
learned more recently about another new way of introducing more entries in the
list of memory types when seen from the endurance point of view.
There
isn't a word for describing this - so let's just call it - software hardened
virtual endurance.
And the unexpected thing is that after this
software hardened virtual endurance figure you can still apply over and above
that another improvement ratio due to other methods.
This means that
SSD designers have another set of memories they can play around with to fit
their DWPD pricing curves. Because for nearly every real physical memory type
they're considering there's an associated and software spawned hardened
endurance which can be upto 10x better.
Where does this come
from? And who's been working on it?
Several companies have been known
to be working on similar techniques. Most notable among these now being NVMdurance whose
technology story is long and convoluted.
I chatted with their CEO
about how their technology fits into the trends of managing SSD endurance and
wrote about it in here -
relating
NVMdurance's machine learning to manual tuning (July 2016). | | |
| ... |
| what about DWPD in consumer
SSDs? - part 1 of 2 - guessing games |
SSDs designed for the
consumer market
typically don't come with 5 year DWPD ratings - partly because that's not the
key characteristic in their selection - but also because almost none of the
companies who design them would be foolish enough to offer a 5 year write
intensive related warranty.
But we can infer a ballpark "consumer
DWPD" rating from published data.
This makes for a cautionary
tale - which can be interpreted in 2 completely different ways.
- it provides some good arguments for avoiding consumer SSDs in write
intensive enterprise arrays
- it shows that consumer SSDs (in 2.5" form factors) are better than
you might have expected - but are still not nearly as good as low end
enterprise SSDs - designed for read intensive cloud applications
Where
does the data come from?
A longtitudinal study based on a set 6
consumer 240GB (SATA) SSDs - from 5 leading companies - called the
the
SSD Endurance Experiment - published by the
Tech Report.
I chose as a convenient reference point the
December
2014 update - at which stage the test had been running for about 14 months -
interspersed with short periods in the powered down state to check data rention.
By
this time 4 of the 6 drives had failed.
If we take the average for the
failed drives - and assume instead they had received the identical terabyte
amount of writes - but spread out at a slower write rate per month - this group
of consumer SSDs failed at about 2 DWPD for 5 years.
This is not the
same as a 2 DWPD rating - because the point of such a rating is to signal to
the designer that most SSDs used in this way would survive - instead of fail.
So
you'd have to downgrade to a lower figure.
But it does give a
comparison point.
Does it matter that 2 of the 6 drives may be better -
and were still operational when this interim update was published?
No -
because as a systems specifier - there's enough data already for you to decide
how you should regard this type of drive.
And endurance is only part of
the story.
Although having said that -
reports from the SSD
data recovery market at the end of 2014 - suggested that endurance related
failure (in consumer and notebook SSDs) accounted for significantly more
failure incidents than all the other component failure causes combined.
...Later:-
the endurance SSD "experiment" data mentioned above above was based on
older SSDs. And as we know from
flash memory basics -
older (in 2D) equates to larger cells and better intrinsic endurance.
In
2016 - the consumer market moved in a direction where you didn't have to play
guessing games any more as you can see in the story below. | | |
| ... |
| what about DWPD in consumer
SSDs? - part 2 - end of guessing games |
Editor:- February 3, 2016 - TLC was originally
intended as a consumer SSD technology (not that you'd realize this from reading
about all the enterprise arrays which have assimilated it).
OCZ recently announced
availability of a 15nm TLC based consumer range of 2.5" SATA SSDs - the
Trion 150.
One of the interesting things about how the marketing of consumer
SSDs has evolved is that these new SSDs come with
DWPD guidance ratings
which are 0.25 DWPD.
Be aware, however, when comparing DWPD ratings
for consumer, enterprise and industrial SSDs that the warranty periods for these
different classes of drives - are different.
The Trion 150
warranty is 3 years -
which is typical for client SSDs - rather than 5 years (as for enterprise
drives).
Endurance
related marketing messages have come a long way in the past 12 years or so.
In October
2014 IBM said (in
effect) "You don't need to worry about the endurance of our FlashSystems."
That was my summary of an IBM blog at the time.
Nowadays OCZ says this
about their Trion SSDs...
"Never Fear, OCZ Endurance is Here."
In
one way I've got to admire the reckless implied simplicity of OCZ's endurance
message. But I also groan in anticipation of how other vendors will retaliate
with similar endurance messages of their own.
I think OCZ's "never
fear" tagline may have been around since last summer (for the earlier
Trion 100 - which OCZ says "quickly became a top seller for us")
but as I don't visit consumer SSD pages any more than I have to (even
my own) I didn't
see it until today.
See also:-
Branding Strategies
in the SSD Market | | |
| ... |
SSD
news
SSD
Reliability
the
Top SSD Companies
SSD endurance -
forever war
can you
trust SSD market data?
bad block
management in flash SSDs
how fast can your SSD
run backwards?
meet Ken - and the
enterprise SSD software event horizon
why is 3D nand
flash endurance better at the same geometry than 2D?
|
| ... |
|
StorageSearch.com is published by
ACSL |
|
 |
endurance needs are application dependent |
| .. |
SSD news
SSD history
storage market research
the problem with
Write IOPS - in flash SSDs
the enterprise
flash story... could it have been simplified? |
| .. |
|
|
| .. |
 |
.. |
|
|
.. |
 |
.. |
| an SSD way
of looking at hard drives |
Editor:- May 4, 2016 - In an ironic twist of fate
- it looks as though hard drive vendors may find it useful to characterize some
aspects of HDDs in a way which can be easily related to value judgement numbers
created for SSDs.
A
recent article -
when
did HDDs get SSD-style DWPD ratings? - in The Register brings to our
attention that hard drrives are now being specified with write limits and the
author Chris Evans
(who also blogs as Architecting IT )
conveniently provides readers with a list of HDD models along with their DWPD
equivalent ratings. ...read
the article | | |
| .. |
| toughening
up DWPD |
Editor:- October 28, 2015 - DWPD ratings have
become a useful shortcut to filter enterprise SSDs because there's an
industry-wide consensus that the number should somehow map into
recognizable
application zones and price bands.
This shows how optimistic the
SSD market mood still is today - when you factor in the jitter level
uncertainty of exactly how elastic that DWPD drive number really is (in the
minds of its creators) and how much it will get twisted around, modified and
stretched when it meets up with the
(is it really) DRAM,
software and SSD
array cousins with which it will cohabit life in the
box.
All
well and good - and my reference article on
DWPD examples in the
market was already becoming quite popular earlier this year (which was no
surprise - as it's simply another way of talking about
endurance)
when I got a wake up email (in April 2015) about a new military grade
industrial SSD which had a DWPD rating.
I didn't write about it at
the time because even though the product had been stealthily working its way
into designs it hadn't been publicly launched. Nevertheless I kept my eyes
open for signs that others might also be doing similar things.
The
similar things being:- the growing use of enterprise architecture in mobile
datacenters and portable and remote rugged systems.
SSDs have been
used in such systems for over 25 years - but often this was essentially a
repackaging exercise to place a rack of industry standard stuff into a
dustproof, ant-fungus treated, drop resistant box with an invertor so it could
run off batteries while keeping the weight and size down so it could be
lifted onto a truck or plane and survive long enough to do useful
data-capture and analysis in the field.
The modern aspirations of
these engineering systems are to do more of the same old things in less
space but also to do entirely new things in widgets which you'll probably see
in season 20 of NCIS.
That's why you're going to see more military
grade, secure, rugged, industrial SSDs coming onto the market with full fledged
DWPD ratings.
It's no longer just an enterprise market parameter.
DWPD rated SSDs are
getting tougher. | | |
| .. |
| SAS SSD
shipments beefed up by DWPD-lite models |
Editor:- June 2, 2015 - Commenting on the
SAS SSD market - Don Jeanette,
VP - TrendFocus
says in his new blog -
SAS
SSDs continue to show strength in an ever increasing competitive market - "SAS
SSDs are not getting squeezed out by the incursion of SATA on the low end and
PCIe on the high end as many have thought would happen."
 Don
explains that SAS SSD makers have populated their product lines with value
models which have much lower DWPD
ratings than the headline performers - which can go some way towards
cost competing with
SATA SSDs while at the same time validating the higher prices of 25 DWPD etc
models. ...read
the article
Editor's comments:-
Rackmount SSD
users don't have to compromise
reliability
when they choose SATA
SSDs in their value engineered arrays.
In a conversation with
Andy Lee,
Marketing Director - EchoStreams
(a whitebox storage company) - in January 2015 - I learned that one of the
design elements in their 20 bay 1U and 48 bay 2U systems - for their customers
who want to build high density flash arrays using COTS SSDs - is that
EchoStreams have deployed
interposers
on their drive bays which allow COTS SATA SSDs to be used in the same way as
dual ported SAS but at lower system cost.
| | |
| .. |
re MLC
how does
NV become V?
NV + 0.4 DWPD @ 85C = V |
Editor:- October 24, 2014 - Even a modest
amount of drive writes per day can render
modern day MLC flash
incapable of retaining data for long in the unpowered state - depending on the
temperature in the rack where those writes took place. This effectively means
that the flash inside the SSD is no longer "non volatile".
The
physics behind this are revealed in a blog by Virtium - a company
which operates in the industrial market - and which does a lot of work
characterizing memories for use in SSDs and other memory systems. They can
leverage that knowledge for customers by adjusting controller and firmware
characteristics to optimize the memory's life and
data integrity -
particularly if it is known in advance what proportion of time the embedded SSD
is likely to be operating at particular temperatures.
Virtium's paper
-
temperature
considerations in SSDs (pdf) includes some stark graphs and observations
about data retention - which you should be aware of - even if you're not in the
industrial market.
Virtium's paper says - "This shows the dramatic effects that
temperature has on data retention for given workloads.
"For the
same 750 full drive writes (0.4 DWPD
drive writes per day for 5
years), SSDs operated and stored at 85C will only have 2 days of data
retention, whereas those drives at 40C will have 1 year and those at room
temperature 25C will exhibit characteristics of nearly 8 years of data
retention." ...read
the article (pdf) | | |
| .. |
Adaptive R/W
flash care & DSP ECC in SSDs
Efficiency - making the
same SSD - with less flash
How will the hard
drive market fare... in an SSD world?
flash wars in the
enterprise - SLC vs eMLC vs MLC vs TLC |
. |
|
|
. |
10 years ago there weren't
so many people who worried enough about what happens inside SSDs to scrutinize
the subject of reliability at this level of detail.
I checked our
stats archive and in in 2005 only 7,500 people read the SSD wear leveling
article here on StorageSearch.com. (That was 1% of our readers at that time).
In those days if an SSD maker said their SSD was reliable - you
believed it. And there weren't so many alternatives to choose from. |
| razzle dazzling flash SSD
cell care and retirement plans | | |
| |
