| Contact us:
editor@storagesearch.com DWPD
fastest SSDs
storage reliability
ratios in SSD architecture
RAID, RAIC, RAIN, RAISE etc
11 Key Symmetries in
SSD design
adaptive R/W and
DSP ECC in flash SSDs |
.. |
| ..... |
|
| classic
SSDs from
SSD market
history |
|
|
| Integrated with multiple
patented and patent pending technologies, the
Immortal™ series solid state drives are designed to work in extreme
environmental conditions in military, defense, aerospace, and submerged
applications
Military grade SSDs from
Foremay | | |
|
.. |
|
1.0" SSDs
1.8" SSDs
2.5" SSDs
3.5" SSDs
1973 - 2018 -
the SSD story
2013 -
SSD market changes
2014
- SSD market changes
2015 - SSD market
changes
2016 -
SSD market changes
2017 - SSD market
changes
2018
- SSD market changes
20K RPM
HDDs? - SSD killed RPM
About the publisher -
1991 to 2018
Adaptive R/W
flash IP + DSP ECC
Acquired SSD
companies
Acquired
storage companies
Advertising
on StorageSearch.com
Analysts - SSD market
Analysts - storage market
Animal Brands in
the storage market
Architecture
- network storage
Articles - SSD
Auto tiering
SSDs
Bad
block management in flash SSDs
Benchmarks - SSD - can
your trust them?
Big market picture of
SSDs
Bookmarks
from SSD leaders
Branding Strategies in
the SSD market
Chips
- storage interface
Chips
- SSD on a chip & DOMs
Click rates - SSD banner
ads
Cloud with
SSDs inside
Consolidation
trends in the enterprise flash market
Consumer SSDs
Controller chips for SSDs
Cost of SSDs
Data recovery for flash
SSDs?
DIMM
wars in the SSD market
Disk sanitizers
DRAM (lots of stories)
DRAM remembers
DWPD - examples from the market
Efficiency - comparing
SSD designs
Encryption
- impacts in notebook SSDs
Endurance - in
flash SSDs
enterprise
flash SSDs history
enterprise
flash array market - segmentation
enterprise SSD
story - plot complications
EOL SSDs - issues for buyers
FITs (failures in time) &
SSDs
Fast purge /
erase SSDs
Fastest
SSDs
Flash Memory
Garbage Collection and
other SSD jargon
Hard drives
High availability
enterprise SSDs
History of data
storage
History
of disk to disk backup
History of the SPARC
systems market
History of SSD
market
Hold
up capacitors in military SSDs
hybrid DIMMs
hybrid drives
hybrid storage arrays
Iceberg
syndrome - SSD capacity you don't see
Imprinting the brain of
the SSD
Industrial
SSDs
Industry trade
associations (ORGs)
InfiniBand
IOPS in flash SSDs
Jargon - flash SSD
Legacy vs New Dynasty -
enterprise SSDs
Limericks about
flash endurance
M.2
SSDs
Market
research (all storage)
Marketing
Views
Memory
Channel SSDs
Memory Defined
Software - yes really
Mice and storage
Military storage
News
Notebook SSDs -
timeline
Optical
storage drives
PATA
SSDs
PCIe SSDs
Petabyte SSD roadmap
Power loss -
sudden in SSDs
Power, Speed and
Strength in SSD brands
PR agencies - storage and
SSD
Processors
in SSD controllers
Rackmount SSDs
RAID systems (incl RAIC RAISE
etc)
RAM
cache ratios in flash SSDs
RAM memory chips
RAM SSDs
RAM SSDs versus Flash
SSDs
Ratios in SSD
design architecture
Razzle
dazzling flash care
Reliability - SSD /
storage
RPM and
hard drive spin speeds
SAS SSDs
SATA SSDs
SCSI SSDs - legacy parallel
Security
Services
Software
Symmetry in SSD design
Tape libraries
Test Equipment
Top 20 SSD companies
Training
Tuning SANs with SSDs
USB storage
User Value
Propositions for SSDs
VCs in SSDs
VCs in storage - 2000 to
2012
Videos -
about SSDs
Zsolt Kerekes -
(editor linkedin) |
|
.. |
|
|
|
|
StorageSearch.com is published by
ACSL founded in
1991.
© 1992 to 2018 all rights reserved.
Editor's
note:- I currently talk to more than 600 makers of SSDs and another 100
or so companies which are closely enmeshed around the SSD ecosphere.
Most
of these SSD companies (but by no means all) are profiled here on the mouse
site.
I still learn about new SSD companies every week, including
many in stealth mode. If you're interested in the growing
big picture of
the SSD market canvass - StorageSearch will help you along the way.
Many SSD company CEOs read our site too - and say they value our thought leading
SSD content - even when we say something that's not always comfortable to hear.
I hope you'll find it it useful too.
Privacy policies.
We
never compile email lists from this web site, not for our own use nor anyone
else's, and we never ask you to log-in to read any of our own content on this
web site. We don't do pop-ups or pop-unders nor blocker ads and we
don't place cookies in your computer. We've been publishing on the web since
1996 and these have always been the principles we adhere to. |
| |
SSD Year 2018?3 things which have already happened and 1 which
hasn't yet
by
Zsolt Kerekes,
editor - StorageSearch.com - November 12, 2018 |
| . |
 |
at this time of
year Megabyte sets aside
more time for reading and re-reading
articles about SSDs | Is there an easy way to summarize
the main developments in the SSD market of 2018 already?
In
past years
such articles have been publishable from around the start of November.
And with a few cosmetic edits those past summaries have generally withstood
the test of time.
But the closing months of the calendar aren't as
non eventful as that suggests - and every now and again - vendors have used the
news snooze of the holidays to perform
acquisitions.
Here are some examples.
- December 2012
- Samsung acquired NVELO (a flash caching software company).
- December 2018 - this hasn't happened yet - and is the "1" which I
referred to in my headline
So on the subject of writing year end
summaries of the SSD market - it's more accurate to say that - generally in
recent years - it has been possible to capture all the major
trends in such articles before all 12 months had elapsed. And looking ahead -
if there isn't an acquisition story in the December 2018 news - then you should
watch out for a note of one having been quietly done in January 2019.
So
- what can we say did happen in SSD Year 2018?
For the essential
month by month details you can see the SSD news archive pages below. |
|
|
But I didn't need to revisit any those pages to
write my shortlist - of 3 things - which goes likes this.
3 things in SSD year 2018
1 - international trade barriersIn 2018 - the free flow of memory and
SSD products between the US and China which had characterized business trends
since the start of the modern era of SSDs - came to an end.
A
growing set of patent disputes, government embargoes and the background
worldwide babble of trade wars (related to other countries - not just
involving the US and China - and related to other types of products too)
created market conditions for memory and SSD companies in which the mood of
doing business in 2019/2020 is expected to be different to how things had worked
before.
Will this affect prices?
We saw the SSD market
survive and thrive when the costs of memory doubled during the
2017
memory shortages. Set against that market robustness and expectations of
falling prices
again (the way they used to be) - then I think tariff barriers won't have
a damaging effect on prices.
Will trade barriers affect SSD company
prospects?
Yes - but it's a mixed picture.
Sme companies
(whose business models leveraged the freedom of micro managing
international differences in the costs of assembling products) will see
initial setbacks and will have to redraw their supply chain maps.
On
the other hand - there will be new opportunities for smaller companies
arising from a trading climate in which there is more local protection from
locked out dominant international competitors..
And companies
which operate outside the disputed hot spots - will see opportunities to
service markets in which they would have been uncompetitive before.
2 - memory cost outlookIn the 4th quarter of 2018 - the imbalance
between supply and demand of traditional memory products (nand flash and DRAM)
which had for 2 years flipped the price per bit curve outlook upwards from
its historically
established downwards direction - looked like it was getting ready to flip back
again - according to price trends being reported by leading
market reporters.
Looked
at in isolation - it would be reasonable to expect that the memory industry's
ability to ship more products in 2019 - than it could when still dealing with
the overhang of yield issues related to the structural switch away from the
last significant generation of 2D scaling - would inevitably lead towards
expectations of lower prices.
However - the impact of trade wars
introduces new variables into the cost outlook for memory systems.
The
interplay between the industry's ability to manufacture traditional memory
products - compared to the growing difficulties of legally shipping them into
traditional geographic markets will create a rich vein of source material for
SSD bloggers in 2019.
3 - memory defined processingBecause
standards organizations
have traditionally taken so long to say anything meaningful about the past
disruptive trends in the modern era of SSDs (such as the gap between the
proprietary PCIe SSD market invasion of the server market and its taming by
the unifying language of NVMe) you know that when a standards ORG
intervenes in this market - then something has really happened.
So in
2018 when SNIA said it was interested in getting involved in talks about taming
the computational storage market - then that was another one of those things.
But
a problem for the industry - which hasn't been solved yet in 2018 - is what to
call this thing.
I 've touched upon it each time a news story has
appeared on these pages and the
SSD jargon page
includes some (but by no means all) the examples which have been used by various
companies which have implemented products.
- in-situ processing
- processing in memory
- computational storage
- in-memory computing (historically means something different - but is
starting to sound like its meaning should change)
This is one of those
subjects - like the original adoption of SSDs - about which we already know a
lot - except what to call it.
It intersects with many other top level
views of memoryfication architecture.
You can describe it as one of
the eleven SSD
design symmetries - specifically - "adaptive intelligence flow
symmetry".
Or you can say it's a variation of the data
industry's creative use of
controllernomics
- whereby any data whose latency is beyond the here and now in this chip -
whether due to media speeds or distance and the speed of light - requires a
local intelligent agent which can do useful things at our remote bidding.
But
the concept still lacks a universally agreed name.
And by "universal"
I mean a term which is usable whether the memory is an SSD or RAM array - and
whether the local intelligence is an SSD controller, FPGA, ASIC or other
acceleration engine.
In my headline above - I used the words "memory
defined processing". I wasn't being provocative in offering that. I don't
think it will stick. But real concepts need acceptible words. And just as
memory
defined software was a temporary placeholder in a blog for a real market
concept - even if the words seem as if they are in the wrong places. I think
that 2018 is the year that the concept of memory defined processing was crying
out for a better name than it had received so far.
See also:-
2018 timeline and
permalink for this article | | |
| .... |
strategic bifurcations in SSD market history4 ways to split SSD
history into "before and after" to understand now
by
Zsolt Kerekes,
editor - StorageSearch.com - October 29, 2018 |
|
| . |
 |
we're going to
add the comments
to the source code later | When
I published the first version of my popular article
Charting the
Rise of the SSD Market in October 2004
it
wasn't the long rambling messily formatted historical narrative which you see
today.
In 2004 I and my readers in the SSD market had already lit the
fuse for the market explosion which I predicted would inevitably follow - based
on a market adoption and "why buy SSDs" user value propositions which
were completely different to anything which the industry had thought about
before.
I was too busy then changing all my plans to reduce my
editorial coverage of all other storage subjects to focus all my energies on
the SSD market. I was more interested in the
future than the
past. And I wondered if I'd be able to keep up with tracking market and making
a difference to it. |
So in 2006 when when I was looking ahead at my
predictions about the growing number of SSD companies which no one had heard of
(and which would need a
new series called
the Top SSD Companies) I realized that SSDs was a market without a reliably
written history.
How would newcomers understand it? And we needed
those newcomers to make the market work. And they as early adopters were
already facing a big barrier of challenging intertwined technical issues in
processors, memory, storage and software which was did not have any clear
centralized ownership.
Due to my past experience with another growing
market before that -
SPARC systems
- the quickest way to write the first draft of history is to repurpose
news stories
you've already written into a simple timeline. So in October 2006 my
charting the rise of the SSD market article was repurposed to look a bit like
the history article today. You can see that version
here.
And
after that quick and dirty first version of SSD market history (based on what I
had experienced first hand) I continued to be busy writing about current
and forward looking stuff. So my history article just became a dumping ground
for adding more stories each year.
It also became a fertile source
from which other publications extracted timelines and sometimes entire clumps
of text. Irritating for me that so few acknowledged what their source was -
but hey that's the internet.
interpeting SSD history on a rolling
basis
Something which every market analyst and editor does at many
times in their working lives (if they're any good is to try and interpret for
their readers how news stories relate to the emerging market context. That's
how we get all those end of year "looking back" and predictive "looking
foward" articles which seasonally become increasingly common as December
draws nearer.
I was doing exactly the same kind of thing and in some
years I would confidently assert "this will be or has been the year of such
and such important emerging trend".
So I added those into my
history article too.
And because I care about authenticity (and because I was simply short
of time) I just cut and pasted those present tense analyses into the growing
narrative - whether they were right or wrong - was another matter.
Which
is how we get to this point here.
As I'm retiring from such
active involvement in the SSD market in 2019 (which may already have happened
by the time you read this) I was looking back at SSD history and asking myself
- is there a story which I can wrap around the past 40 years of rambling
anecdotes and an interpretation which might help a modern reader to appreciate
what happened - without having to know all the details?
That was my
eureka moment - when I realized - I've been doing the same kind of thing to
explain SSD controller
architectures, and
memory caching
ratios and design
symmetries etc for many years. And technology experts who know far more
than me about what goes on inside semiconductors and systems have told me they
found my simple splits of architecture into different sets have been handy ways
of thinking about stuff.
So how to split SSD history?
Here
are my 4 proposed strategic splits. |
|
split 1
before and after the Modern Era of SSDsBefore the Modern Era - the
enterprise SSD market wasn't a sticky market.
After the Modern Era -
it was.
And the only thing likely to replace an enterprise SSD was
another SSD or memory system.
In my article
celebrating the
first decade of the Modern Era of SSDs I placed the tipping point at around
2003.
Before the Modern Era - even if an end user had deployed
SSD accelerators to fix performance problems in one key application the balance
of probability was that for future applications the users would turn back again
to legacy solutions such as faster clocked servers and storage. But those
legacy options stopped getting faster in a series of steps:-
- server processor makers had begun telling me in 2000 that future clock
speeds couldn't keep growing in the same kind of way which had been an
associated assumption of semiconductor design shrinks and Moore's Law since the
start of the microprocessor era.
For processors this aspect of
Moore's Law (incrementally faster clock speeds) broke in 2000 to 2003.
I
wrote about this problem in a spoof article April 1, 2002 -
HotServer
Technologies Announces the 3GHz hotSPARC 9000 and more strogly later -
Why Sun Should
Acquire an SSD Company (May 2004) in the SPARC Product Directory.
I
didn't realize in 2000 that clock speeds would stay at the same kind of speed
limits for the next 20 years.
But the good thing about the failure
of these legacy markets to deliver ever faster solutions was it forced more
people to look at SSDs - despite the steep initial learning curves involved.
|
|
split 2
before and after Fusion-ioIn
September 2007 -
Fusion-io launched
the ioDrive - a PCIe form factor flash SSD with upto 640GB capacity and 100K
IOPS performance.
That event created the
PCIe SSD market which
soon after became a key focal point for innovations in enterprise SSD
architecture and SSD systems software due to the combined efforts of the many
competitors which entered that market.
There are many strands of
SSD market history associated with Fusion-io. In this article about strategic
bifurcations I'm just going to mention 2 of them.
SSDs became
must-have options in server product lines
Before 2009 -
when Fusion-io began announcing a series of design wins for its ioDrives in
enterprise servers made by HP, IBM and Dell - big server manufacturers didn't
offer SSD accelerators as standard options in their product ranges.
After
these initial design wins - it became
imperative
for all server manufacturers to offer an SSD solution embedded in future
products if they wanted them to look competitive. (This tipping point - if one
does it they'll all have to do it - was predicted in my 2003 article -
could enterprise SSDs become
a $10 Billion Market?)
the strategic importance of SSD software
Before
Fusion-io there wasn't an SSD software ecosystem. And the lack of automatic
software tools for integrating SSDs into caching and other acceleration roles
meant that deploying SSDs as accelerators required expensive and
skilled manual hot
spot tuning.
One of the spin offs from the standardization of SSD
integration in servers was that it inspired a wave of
SSD software startups
which had never been viable before.
Most of the early SSD software
pioneers were acquired
by SSD manufacturers as the market learned that having compatible software for
virtualization and caching
made it much easier to sell their SSDs.
Although it's hard to
believe it now - this was still a time when there was a
vacuum in the
space where the SSD software brain might have been expected to be in the
vision of legacy giants in the enterprise software, processor chip and array
storage markets - who were all taken completely by surprise by this new
industry. |
|
split 3
before and after Diablo Technologies' Memory1 Diablo Technologies was a
pioneer in shipping SSDs and DRAM emulation products which consisted of flash
memory and controller IP integrated in a DRAM compatible DDR3 or DDR4 bus slot.
Although Diablo's products failed to achieve any significant market
traction and the company was involved in a series of patent disputes and went
bankrupt - there were important market lessons to be learned from the outcome of
what I called in August 2015 the first salvo of
SCM (Storage
Class Memory) DIMM Wars.
Before Diablo's Memory1 (announcement,
shipping and customer benchmarks) there had been a genuine belief in the SSD
industry that new types of non volatile memory memory products shipped in DRAM
compatible bus slots would plausibly launch the next wave of performance
improvements in the server market - in a similar way that the PCIe SSDs had
done earlier - and that somehow - the latency differences of these 2 types of
busses and the ability to place more emulated memory in server motherboards
would make an order of magnitude difference and help to create substantial new
markets.
The immediate effect of the Memory1 announcement was to set
off a spate of competitor announcements about NVDIMM based products aiming at
the same idea but based on a variety of design approaches and memory
technologies (not just flash).
After the market failure of Memory1 my
analysis of the SCM DIMM wars phenomenon is that the architectural promise of
such products was fatally flawed and delivered marginal incremental benefits
sometimes rather than sustainable order of magnitude improvements of the types
promised by the initial hype.
The main reasons for the past failure and
future limitations of such products are (in my view) the misconception that a
single component type of solution is in itself enough to take computing
performance to a sufficiently high next level - compared to the comparison point
of an already sophisticated SSD infrastructure.
Instead - what I call
the "memoryfication" of enterprise computing - at the next level of
performance will require changes in processor, memory and software architectures
working together in new conceptual schemes.
And I think such
solutions will be agnostic with respect to form factors and may not indeed look
anything like a DIMM. Indeed they may work just as well delivered in PCIe slots
(like Google's TPU) or be implemented at rack scale in fabrics.
I've
discussed these ideas in a bunch of articles including:-
optimizing
CPUs for use with SSDs
introducing
Memory Defined Software
RAM
in the post modernist SSD and Memoryfication Era
should we expect
more from memory? - after AFA's what's next
are we ready for
infinitely faster RAM? (what would it be worth)
|
|
split 4
before and after the memory shortages in 2017Although the silicon
based semiconductor memory industry had experienced
many
periods of under supply and over supply capacity since its formation - the
shortages of nand flash and DRAM which began in the 2nd half of 2016 and lasted
through to the first half of 2018 were like no other which had happened before
- in strategic impact and lasting consequences.
Before the memory
shortages of 2017 - there had been a bunch of non volatile memories which (in
some cases for more than
a decade) had failed to ever emerge into sticky design wins and market
adoption from their seemingly never ending "emerging" status - due
to their lagging too many years behind successive improvements in mainstream
memory pricing per bit.
During the memory shortages - the effect of
rising prices in flash and DRAM and the realization that those legacy
memory roadmaps could no longer be relied on to follow the historic
expectations set by past experiences were beneficial to the competitive
comparisons with emerging competing nvms. It was as if the emerging nvms had
got into a time machine and emerged looking 2-4 years better.
After the
memory shortages of 2017 - it became clear that (as I phrased it in
a 2017 article)
"new notes had been added to the music of memory tiering. And while we
still can't be sure which of these "no longer emerging but emerged"
memories will have lasting power in the memory and SSD markets - it is clear
that a significant change had occurred and (for a forseeable bunch of years)
there is no going back to the just 2 main types of memory model for future SSD
designers and memoryfication architects. Memory designs will change. Processor
designs will change. Architecture will change to incorporate the capabilities of
new types of memory.
A much longer list of market impacts can be seen
in my article -
miscellaneous
consequences of the 2017 memory shortages.
The causes and
analysis of market events leading up to the shortages along with
contemporaneous commentaries can be seen in the
news archives of
StorageSearch.com and other historic web sites which discussed the memory
and SSD markets at that time. |
|
split 5, 6, 7, 8, 9, 10...
insert your own hereI'm sure that many of you having reached this
point may be wondering - why didn't I mention a particular thing which you
believe is just as important as those above.
I'd like to think that
having written so many articles about SSDs - maybe I already did write about
some of those other things before. But if I didn't - then I'm sorry. I'm glad I
finished this one. It's the last main article about SSDs I'll be writing about
SSDs this side of retirement. (Or maybe ever.)
Thanks for reading it.
And if it's given you some ideas - then that's what I was hoping.
Here's
a very old article I feel proud to have written.
why buy SSDs? -
pioneering use value propositions
Bye for now and take care. | | |
.... |
 |
.... |
|
by
Zsolt Kerekes,
editor - StorageSearch.com - September 14, 2018 |
What's the connection between
fast erase SSDs,
storage drive
sanitization and data
recovery?
It was the recent news about the hurricane season
which got me thinking again (as it does every year) about all those tragic
stories of damage and loss.
The data recovery industry is unlike other
parts of the computer market because no one sets out with the idea that they
want to include DR technology in their annual storage budgets.
I
thought back to some of the stories of amazing data survival and recovery which
I've reported on in my writings here.
Drives being soaked in mud
(that's one thing) - drives being drenched in chlorine (including the backups)
that was another. |
.. |
 |
| we know who
can fix it | | |
And similar to the concept in the days when spy
satellites intentionally crash landed their images on film cannisters in the
ocean, there have been science projects in the digital camera and telemetry age
which were designed to bring back space data from crashed balloons in SSDs.
Sometimes bad things happened to the SSD afterwards. I had forgotten about
but still love this
SSD
data survival story about Memtech - from 2005.
Re data loss -
human error plays a big part in many stories of data loss (even in replicated
servers) while human
error is also one of the biggest vulnerabilities in data security.
In a reflective article looking back at my conversations with the data
recovery market I note some of the things which surprised and impressed me at
the time.
And having tracked the SSD data recovery market since before
it began - I found a way to connect all the dots. ...read the
article | | |
.. |
20 years ago - in stealth mode by
Zsolt Kerekes,
editor - StorageSearch.com - September 12, 2018 |
I 've been perusing my emails of 20
years ago (September 1998) which was the month before I was due to
launch my new site
StorageSearch.com and I was curious
to remind myself what kind of conversations I was having with people in the
run up to doing that and - in effect - exiting stealth mode with an entirely
web based publication which had no past print rooted brand strength.
I had already been publishing a server market guide for 6 years at that time
(called the
SPARC Product Directory)
and my company had
been operating a dotcom based business model funded by web advertising
since 1996. But I wanted to make the new storage site look different and have
its own identity.
Some of the new changes I planned to introduce in
my new server-agnostic storage guide were more pictures (for example the
branding images of the
first few mice were already in hand) and the other "innovation"
was to be the use of banner ads on the site - a design feature to which I had
originally said no a few years before. |
.. |
 |
when dealing with such a small
cake
practicality dictates that one candle
for each decade is quite
enough | | |
In those 20 year old emails I was looking at
this week a common thread was that none of my storage advertising customers
had ever run banner ads to promote their enterprise products before and so
therefore many of the conversations were about finding people to design
them in time to launch the new site.
I have preserved some of these
older ads for future historians (pre modern SSD era
storage banner ads /
pre
millenium SPARC banner ads) because they still tell us a lot about the
products and the excitement about the key messages about them from a
perspective which you don't see in dry historical narratives. (I'll add some
more of those early examples into an update of this blog so you can get an idea
of how much things have changed since then.)
You can see below an
example of one of the earliest storage banner ads designed specifically to run
here.(The vendor was Dynamic Computer Products which soon after changed their
business name to
Data
Storage Depot. |
 |
This rev 1.0 ad design above - "leaders in
RAID, disk, tape and memory upgrades" - was created in September 1998 and
like most ads underwent many design changes during the period that the ad
programs ran. (Approximately 8 consecutive years with this customer - which was
not unusual in my customer base during the 20 peak years of my web ad based
business.)
If it wasn't for the brave companies who were designing ads
for a new site which didn't exist yet and who weren't scared of mice then I
wouldn't have had such a rewarding job for the next 20 years and my readers
would've had to wait another 2 years to see the next wave of
portals (that's what we
called them back then) which covered the storage market as a disaggregated
whole.
StorageSearch.com's initial
storage-wide focus
(from raw storage chips, and RAID systems,
backup software
to tape libraries) itself
would later change when in about 2007 - on seeing the scale of my
earlier predictions
for the SSD market coming true - I quietly resolved to reduce my
editorial coverage of things to do with the rotating storage market and instead
refocus most of my energies on the rapidly changing SSD market. Which was
just about as much as I thought I would be able to comfortably wrestle with
given where the technology had started and where I thought it was going to go.
See
also:- SPARC
history, SSD
market history, D2d (disk
to disk backup) history | | |
.. |
3d nand and new dimensions in SSD controller architecture
research exploits layer based differences by
Zsolt Kerekes,
editor - StorageSearch.com - August 28, 2018 |
In the early years of
nand flash memory
adoption in the enterprise (for simplicity let's call this period the MLC
(pre-TLC) / pre-adaptive R/W / pre-DWPD era) there wasn't the
same kind of established delineation of
application
roles for new SSDs as there is now -
 because
SSDs were still carving out new reasons to be used in design wins (almost one
startup at a time) and it also happened quite often that when a new product
was announced there would be significant gaps in the datasheets compared to what
was needed to be known to determine how the product might behave (without having
to invest large amounts of resources into benchmarking and evaluations).
To
help my readers in this formative period I suggested
several shortcuts
which could help potential integrators group such new SSDs into sets determined
by key design and architectural decisions in the new SSDs.
These
enabled anyone who thought a lot about SSD controllers to decide for themselves
- yes this new one is in this set and so some its characteristics are
preordained - it's better at this, worse at at that - irrespective of whether
there were any datasheets or benchmarks or
whether we believed
that such benchmarking had been correctly set up (which for a long time it
wasn't). I know from the conversations I had with many systems designers that
they found some of my "filtering" terms to be useful shortcuts - and
most of the companies which were creating these new products found it useful to
answer my questions about the internals of their designs and thinking.
But
all such rules of thumb have a limited shelf life. And as I used to remind
readers in my year end articles - it's just as important to discard old ideas
which at one time were useful as it is to adopt new ones.
One of the
simplest SSD design filters which I wrote about was something I called the
difference between
big and small
SSD controller architecture (2011).
At the heart of this was the
question - how many memory chips has the controller been optimized for?
Because if it can work with a single digit set of chips then the controller
can't employ as many clever strategies (to help reliability, performance and
quality of performance) as another design which has been designed with a floor
level of tens or hundreds of chips. It was a simple idea and it was a useful way
to look at controller designs over a 10 year period.
But a paper I saw
this month made me reconsider whether that division still works. And even to ask
the question - are there any small architectue SSDs left at all?
The
paper in question was -
Improving
3D NAND Flash Memory Lifetime by Tolerating Early Retention Loss and Process
Variation (pdf) by Yixin Luo and Saugata Ghose (Carnegie Mellon
University), Yu Cai (SK Hynix), Erich F. Haratsch (Seagate Technology) and
Onur Mutlu (ETH Zürich) - which was presented at the SIGMETRICS
conference in June 2018.
This paper - among other things - suggests several new (not previously
publicly written about) design approaches which can be adopted with tall (30
layers upwards) 3D nand flash - which can leverage characterization assessments
which are made on a small sample of cells in a memory chip and leverage those
with architectural support in an SSD controller to increase SSD reliability or
performance so as to make enterprise use of such memories more attractive.
One
of the ideas discussed in the above paper is the idea that the quality of cells
varies in each layer. This in itself is not new. What is new however is that
the authors show how the spread of reliability can be measured, modeled and
harvested.
The authors say - "We are the first to provide detailed
experimental characterization results of layer-to-layer process variation in
real flash devices in open literature. Our results show that the raw bit
error rate in the middle layer can be 6x the raw bit error rate in
the top layer."
Among the many chip dependent design approaches
in the paper here are 2 which I've singled out.
- LaVAR - Layer Variation Aware Reading
- LI-RAID - Layer-Interleaved RAID
Layer Variation Aware Reading (LaVAR) - "reduces process
variation by fine-tuning the read reference voltage independently for each
layer."
This idea - which properly occurs in the realm of
adaptive R/W
technology (rather than big controller architecture) suggests a simple
model which can predict a best guess threshold voltage for P/E based on a
top/bottom samples extracted after endurance conditioning a small number of
blocks in the memory.
On its own - this concept would be enough to
make the paper a must-read for controller designers.
My gut feel is
this points the way to a middle course of run time controller design between 2
well known philosophies:-
- the adaptive DSP ECC approach - which combines chip learned
characterization with heavy weight run time processing power in the target
controller and
- the machine learning / lifetime based characterization models proposed by
NVMdurance in 2013 -
which enables lightweight run time processing - based on a model which
extrapolates the best figures for a population of all memory chips - but is
learned from a factory based characterization (rather than learned from the
local chips attached to the controller).
Layer-Interleaved RedundantArray of Independent Disks (LI-RAID) - "improves
reliability by changing how pages are grouped under the RAID error recovery
technique. LI-RAID uses information about layer-to-layer process variation to
reduce the likelihood that the RAID recovery of a group could fail significantly
earlier during the flash lifetime than the recovery of other groups."
This
- to me - starts to look like another "big controller" architecture
idea - but the authors say it can be used in an SSD with just a couple of chips.
They also extend the concept to pairing the best predicted blocks in one memory
chip with the worst predicted blocks in another memory chip in the same SSD.
You
can read about earlier uses of RAID thinking in SSD controller designs
(including variable size planes) in my
RAID systems page.
But
it's clear that the interpretation of different layers in a 30 to 100 layer or
so 3D memory chip starts to look a lot like big controller architecture.
Previously
it was the number of different identifiable conceptual toys in the box which
set the limits to system level design tricks. Now it's layers in the same chip
too. | | |
.. |
some earlier home page blogs
re RATIOs in SSD architecture
40 years of thinking
about non volatile memory endurance
miscellaneous
consequences of the 2017 memory shortages
are we ready for
infinitely faster RAM? (and what would it be worth)
introducing
Memory Defined Software - yes seriously - these words are in the right order
|
.. |
| Hmm... it looks like you're seriously
interested in SSDs. So please bookmark this page and come back again soon. |
.. |
 |
| |
.... |
 |
|
see
news page |
| .... |
|
|
| .. |
|
|
| .. |
Since the early
1970s there have been 3 revolutionary disruptive influences in the
electronics and computing markets.
- the microprocessor
- the commercialization of the internet
- the advancement of computer architecture
enabled by the
modern era of SSDs
|
| comparing the SSD
market to earlier tech disruptions | | |
| .. |
| Diskful Writes Per Day
began as a shortcut to describe the endurance of SSDs in enterprise SSDs but
within a couple of years it became adopted by other markets too - for
industrial, military and even (surprisingly) consumer drives. It has been a
useful metric but has its limitations too. |
| what's
the state of DWPD? | | |
| .. |
|
|
| .. |
|
|
| .. |
|
|
| .. |
It was the best of times.
It was the worst of times.
2017 was a year like no other in 40 years
of SSD history. |
| which way next for SSD? | | |
.. |
|
|
.. |
| Despite many revolutionary
changes in memory systems design and SSD adoption in the past decade we are
still not at the stage where it's possible to predict and plot the next decade
as merely an incremental set of refinements of what we've got now. |
| Are we
there yet? - 40 years of thinking about SSDs | | |
.. |
|
|
.. |
|
|
.. |
|
|
.. |
| Enterprise DRAM has the
same latency now (or worse) than in 2000. The CPU-DRAM-HDD oligopoly
optimized DRAM for a different set of assumptions than we have today in the
post modern SSD era. |
| latency loving
reasons for fading out DRAM | | |
.... |
|
|
.... |
How do we know anything?
And how confident can we be when using that knowledge as the basis to
make important decisions?
I'm not talking here "cogito ergo sum" but the rather more
down to earth matter of - how well can anyone today understand the SSD market? -
and give you a reliable answer to a simple question like - what's the best way
of getting to SSD street from wherever your starting point happens to be right
now. |
| Can you tell me the best
way to get to SSD Street? | | |
.. |
|
.. |
|
|
.. |
|
|
.. |
|
|
.... |
| I said to Rob Peglar - "The
ratio of processor cores to memory channels and local memory capacity is a solid
pivot from which to leverage your forthcoming architecture blogs. I love ratios
as they have always provided a simple way to communicate with readers the design
choices in products which tell a lot to other experts in that field." |
| re RATIOs in SSD architecture | | |
.... |
|
|
.. |
|
|
.. |
|
|
.. |
| A couple of years ago - if
you were a big company wanting to get into the SSD market by an acquisition or
strategic investment then a budget somewhere between $500 million and $1 billion
would have seemed like plenty. |
| VCs in SSDs and storage | | |
.. |
|
|
.. |
|
|
.. |
| With hundreds of patents
already pending in this topic there's a high probability that the SSD vendor
won't give you the details. It's enough to get the general idea.
|
| Adaptive flash
R/W and DSP ECC IP in SSDs | | |
.. |
|
|
.. |
|
|
.. |
|
|
.. |
|
|
.. |
|
|
.. |
| Why can't SSD's true
believers agree upon a single coherent vision for the future of solid state
storage? (They never did.) |
| the SSD Heresies. | | |
.. |
|
|
.. |
| If you spend a lot of your
time analyzing the performance characteristics and limitations of flash SSDs -
this article will help you to easily predict the characteristics of any new SSDs
you encounter - by leveraging the knowledge you already have. |
| flash SSD performance
characteristics and limitations | | |
.. |
| The memory chip count
ceiling around which the SSD controller IP is optimized - predetermines the
efficiency of achieving system-wide goals like cost, performance and
reliability. |
| size matters in
SSD controller architecture | | |
.. |
| A popular fad in selling
flash SSDs is life assurance and health care claims as in - my flash SSD
controller care scheme is 100x better (than all the rest). |
| razzle dazzling flash SSD
cell care | | |
.. |
|
| |
