| leading the way to the
new storage frontier||...|
|the SSD Heresies
Top SSD Companies - 2007 to
fast can your SSD run backwards?
are we ready for
infinitely faster RAM?
40 years of thinking
about nvm endurance
not to compile a list of military SSD companies
sanitizing sensitive SSDs? heed my words oh prideful SSD
dust you were born
and unto dust you shall return
|Editor:- September 25, 2018 - A reader - Simon Zola
- Manager, AVTEL Data
Destruction emailed me last week after seeing my recent home page blog -
back at my 19 years of writing about the data recovery market - which I
concluded with this.. |
there an opposite concept to data recovery?
The flip side to data recovery is fast purge SSDs and
Simon said - "I have only just come across you
and your site and I would love to hear your opinion on meaningful data
sanitisation of SSD."
thought to myself how many years is it since I set up a dedicated
SSD fast erase / purge
page? - I checked. It started in 2009. (This is one of the joys and
frustrations of the web. Frustration - that you can't find stuff which has
been around for a long time - because it gets drowned by social chit chat. Joy -
in knowing that there must be a lot more readers out there who also care about
the same problems.)
Anyway - what I said to Simon was - "There is
a double digit list of standards by defence and government agencies which cover
various use cases and whether the drive is desired to be redeployed for another
project or not. The purpose of extreme autonomous SSD purge is to destroy enough
critical chips in the encrypted SSD so that if it falls into the wrong hands
(captured by enemy) then the SSD data will remain immune to the best
efforts of forensic data
recovery. Thats just one reason why DR and security agencies intersect and
are mutually aware. But as DR gets better then sanitisation has to advance too
(best way being destruction of the chips)."
Anyway Simon - whose
company does Onsite Physical Destruction of HDDs and SSDs in Australia pointed
me towards an interesting
video - re Mobile Data
Destruction which shows the type of thing his company does. It's on
youtube which means that many of you won't be able to see it right now if
you're viewing this at work.
So I'll describe what happens...
The video shows a van which arrives at your site and delivers via a conveyor
belt all the drives you want shredded - presumably while one of your security
people watches it happening. (You'd have to verify the exact design and chutes
etc yourself obviously to satisfy yourself there are no magical trap doors - or
maybe you could just rent the facility. It depends on your own circumstances.)
That prompted me to realize that it had been about 2005 when I had
last written much about the
services and equipment business (as opposed to autonomous drive purge)
because in a way - once you know what needs to be done - what more can you say
about it? But maybe that page could do with a refresh - which is why I'm writing
We are much more sensitive and vigilant about environmental
impacts nowadays (2018) compared to the start of my own career (1977) when
many of the industries which paid the wages of our local communities and
where our friends and neighbors and customers worked were inevitably sometimes
spilling stuff into the sky, ground and water.
So I said - Hi
Simon - I forgot to ask this... how is the shredded material from the sanitized
drives processed? I mean the cost from an environmental hazard point of view?
Simon said - 0 to land fill. (And then he gave me a list of who
reprocesses what afterwards - which you can find out more about on his web
Editor's comments:- I'm guessing that wherever you live you
might be interested in the possibilities opened up a mobile service like this.
My own modest needs in this category have always been simply
managed by the expedient of a log splitter or ax - but I'm only smashing one
drive each season or less. Some of the kids of family friends have made
artworks out of the little chunks of smashed up drives and mangled chips.
dustry grains are less artistic but better from the security angle.
|SSD news in
Septembers of yore |
- September 2000
- M-Systems' Diskonchip SSDs appeared on Linux SBCs made by VMIC.
- September 2005
- SimpleTech launched the world's first dual interface SSD.
- September 2007
- Texas Memory Systems began shipments of the
RamSan-500 - a 4U FC or
IB flash based SSD accelerator with 2TB capacity and hot swappable flash
modules. This model from the world's oldest enterprise SSD company heralded
a pivotal market
switch from RAM to flash in performance optimized storage boxes.
- September 2011
- Kaminario announced it was using Fusion-io's PCIe SSDs as a new option in
its K2 rackmount SSD product line. Before that the K2 had been RAM only.
- September 2013
- Micron began sampling the first implementation of its Hybrid Memory Cube
- a DRAM architecture concept - which had been launched 2 years before.
- September 2016
- Nimbus said that
276,480 nand flash dies were used in its recently demonstrated 4.5 petabytes
(raw) 4U ExaFlash storage array.
- September 2017
- UPMEM announced series A funding for its Processing In-Memory technology
which integrates user-API accessible RISC processors as SoCs in DRAM.
|If you could go back in
time and take with you a factory full of modern memory chips and SSDs
(along with backwards compatible adapters) what real impact would that have?
|are we ready for
infinitely faster RAM?
/ more pages
reports on the success of its switch to software business model - and an
investment from Western Digital|
Editor:- October 9, 2018 - Kaminario (which had
been a thought leading
rackmount SSD company
in January 2018
it was no longer supplying hardware systems directly itself) today announced
that it grew its topline software business more than 55% during the first 9
months of 2018, relative to the same time period in 2017.
addition, Kaminario said it has been operating with positive cash flow for the
quarter ending Sept. 30. Kaminario also announced that it received a strategic
investment from Western Digital Capital, the strategic investment fund of
Western Digital Corp.
Editor's comments:- Kaminario's change of
business model was in line with the market adjustment trends I had predicted
in my 2015 article -
mechanisms and routes towards consolidation in the enterprise SSD market along
with some other outrageous and dangerous ideas. Its software can now be
viewed as a fault tolerant platform for SSDs supplied by Western Digital.
This was a pragmatic affinity after a long journey in flash which
may have had some original germ of a root in a conversation I had with
Kaminario's CEO in March 2011 - when Kaminario (which had emerged from
stealth as a RAM SSD
company) was looking for a flash partner. Kaminario - which had always
regarded its software as being the unique branding ingredient in the IP mix of
its systems - grew closer over many years of relationships with a succession
of leading enterprise SSD companies which were later serially
absorbed within WD.
DRAMeXchange says - supercycle of DRAM price hikes is over
October 9, 2018 - DRAMeXchange
how it's interpreting memory pricing and supply trends.
- DRAMeXchange says:-
- DRAM products have begun to see a weak price trend, showing only a 1~2% QoQ
hike in contract prices for 3Q18 due to the continued oversupply, despite the
coming of holiday sales season. DRAMeXchange expects the quotations of DRAM
products to decline by 5% or more QoQ in 4Q18, terminating the super cycle of
price growth for 9 consecutive quarters.
- DRAM manufacturers all expect a high possibility of oversupply in 2019.
Therefore, they have tried to postpone or slowdown the capital expenditure and
re nand flash - DRAMeXchange says:-
- For 2019, DRAMeXchange expects the annual bit output to increase by nearly
- nand flash experienced a price drop of around 10% in 3Q18 and expects a
steeper drop of 10~15% in the fourth quarter, considering the impacts of trade
war. Contract prices of 3D TLC NAND Flash chips in the channel market may even
drop by more than 15% in 4Q18.
- The nand flash market is influenced by the sluggish demand for consumer
electronics, while demand for the more profitable Enterprise SSD from servers
and data centers remains stable. However, the competition among Enterprise SSD
suppliers will become increasingly fierce; hence the prices of Enterprise SSD
are very likely to continue decreasing in 2019.
See also:- storage
market research directory
- On the supply side, nand flash suppliers have raised their output forecasts
as they have expanded their production capacity and improved the yield rates of
their 64/72-layer 3D NAND production.
Clarifying SSD Pricing
- where does all the money go?
guide to semiconductor memory boom-bust cycles
Silicon Motion ships > 750 million NAND controllers / year
October 5, 2018 - Silicon
Motion says "We ship over 750 million NAND controllers annually
and have shipped over 5 billion NAND controllers in the last 10 years, more than
any other company in the world."
They might have been saying that
for some while but I only noticed it today when looking in the footnotes of
2018 preliminary press release which warned that "revenue is expected
to be within the lower half of the original guidance range of $136.0 million to
$142.9 million that the company issued on August 1, 2018." (Maybe that's
what happens if markets adjust to a smaller supply of more expensive than
anticipated memory chips - we'll have to wait to see Si Motion's analysis
on October 30, 2018.)
Editor's comments:- the shipment numbers for
controllers show how large the SSD market has become.
new website for Xccela Consortium - developing new local bus for
embedded memoryfication chips
Editor:- October 4, 2018 - A new
website to support a new storage
today. The Xccela Consortium (at
www.xccela.org) now has 12 member companies working to promote the Xccela Bus as
an open-standard digital interconnect and data communications bus suitable for
volatile and nonvolatile memories as well as other types of ICs.
the technology page..."In
its first iteration, the Xccela Bus is a high-speed, high-performance Octal SPI
bus that uses 8 data lines for command and data transfer. It is fully compliant
with the JEDEC xSPI standard. The bus is synchronous and supports both
single-data rate (SDR) operation, where one byte of data is transferred every
clock cycle, and dual-data rate (DDR) operation in which two bytes of data are
transferred every clock cycle. The Xccela Bus supports clock frequencies up to
200MHz and data transfer rates up to 400MB/sec (3.2Gbps).
bold claims are back from the newly confident sounding
Editor:- October 3, 2018 - Violin made a huge
splash with the readers of StorageSearch.com 11 years ago when they entered the
the Top 10 SSD OEMs
list - 2007 Q3 soon after exiting stealth mode.
By June 2011-
Violin Memory (as it was then called) was confidently talking about being on
its way to building a billion dollar company in this once simple market (when
fastest was a
proposition and SAN had been a good enough business plan (FC SAN
customers had money) until SSD accelerators from
talked their way into server slots with
PCIe in 2007/8) and
even after that there were plenty of
opportunities for rackmount enterprise flash but it was not the same simple
market it had once been.
The enterprise SSD market was a business I
knew well and its strategic nature and size attracted huge numbers of
competitors offering more software bundling options, latency bands and
price points than any single technology platform or company could
As Violin progressed to IPO in
August 2013 its
earlier arguably plausible leadership advantages had already eroded and it
was clear that wanting it all wasn't the same as getting it. This was a market
which was moving towards a
customer rather than technology dictated model where it was better to be
excellent at one thing than above average at too many. In the 3 years after
its IPO Violin slid towards the cliffs of being acquired or bankrupt - the
latter case occurring in
after all the words I had already written about those
bubble years of
rackmount SSDs I've been cautious about writing overmuch about the new Violin Systems too-soon.
I have noted more caution and more measured tones of outlook on their web site -
even when I haven't written about them here.
This week I saw that
Violin's old confidence is back - in a press release about its new boxes with
NVMe over 32Gb FC - and the headline -
fastest enterprise storage in the world just got faster!
made me smile.
It's good to see such confidence emanating from this
direction once more.
And another reason I smiled was because in 2002
I was running ads for a company called
Texas Memory Systems
to promote - "the World's Fastest Storage" - which at that time was
the RamSan-210 .
Violin rev 1.0 emerged in 2007 and for the next several years afterwards they
and TMS were the 2 companies which dominated the fastest SAN compatible
rackmount flash SSD segment.
The competitive swot blurred later when
TMS entered the PCIe SSD market (which Violin wasn't engaged in early enough to
leave a mark) and the 2 companies were less often head to head when Violin
defocused from a latency-centric offering and began bundling software as a
feature into its boxes. (TMS had long made a virtue out of its non server
rooted silicon rich
controller architecture - and even when it was later acquired by IBM - in
October 2012. - its RamSan boxes grew accretions of limpet-like surrounding
software boxes as there was no natural roosting perch for the software to do
anything inside the original boxes.)
Going back to Violin rev 2.0 and "the
fastest enterprise storage in the world"...
perspective it's not clear to me that you could get a concensus of expert
opinion as to what constitutes "fastest enterprise storage"
and what shape, interfaces and capabilities and applications value it should
have. But I like the idea that "fastest" claims are still hard
currency in SSD marketing.
after AFAs -
value of infinitely faster memory systems?
2 new reports on the SSD market
30, 2018 - Forward Insights
has published 2 new reports related to the SSD market:-
storage market researchers
- Storage Class
Memories - "Since the announcement of 3D XPoint memory by Intel and
Micron, increased attention has been focused on alternative memory
who's who in the SSD
market in China?
SSDs - 2013 to 2017 (timeline of stories)
data integrity in DRAM
new paper lists and describes
Editor:- September 19, 2018 -
Survey of Techniques for Improving Error-Resilience of DRAM is a new
research paper by Sparsh Mittal
- Assistant Professor at IIT Hyderabad and Maruthi S Inukonda
published in the
of Systems Architecture.
The authors say - "Aggressive
process scaling and increasing demands of performance/cost efficiency have
exacerbated the incidences and impact of errors in DRAM systems. Due to this,
improvements in DRAM reliability has received significant attention in recent
years from both academia and industry. In this paper, we present a survey of
techniques for improving reliability of DRAM-based main memory. We classify the
works based on key parameters to emphasize their similarities and differences.
This paper is expected to be useful for computer architects, chip-designers and
researchers in the area of memory/system-reliability. ...read
Editor's comments:- Among other things this
paper has a detailed analysis of multi-dimensional ECC schemes, multi pin
correction codes (discussion of fault coverage and overheads), the efficacy
of retirement schemes for pins and chips,
RAID and various schemes
for stacked DRAM arrays.
DRAM news in an SSD context,
in the news archive,
enterprise performance reports and the SSD marketers toolkit
Technologies acquires Demartek
Editor:- September 17, 2018 - There
was a time in the early
adoption years of the enterprise flash SSD / AFA market when due to the
generally poor understanding of flash by traditional rotating storage
benchmarkers and consequential
meaningless benchmarks littering the web pages of computer publications
which had recently become aware of
of SSDs it meant that one of the business factors which could
delay the launch of products for SSD startups when their
prototypes were ready for evaluation was the difficulty in finding an external
test lab which had credibility with independent observers.
independent observers on the web - who needed to see plausible reports -
included the evaluation teams in systems companies who were seeking new
products to shortlist for their own inhouse tests from a disruptive market in
which for many years no one knew where the next great products might be
coming from. I was often asked by SSD startups if I could recommend such
benchmarking report writers. And I talked to customer evaluation teams too.
One of the small number of credible test labs whose published work in this
market I grew to respect in this category was Demartek
Well - this week I saw the
that Demartek has been acquired by a company called
comments:- I hadn't heard of Principled Technologies before but looking at their
portfolio it includes a bunch of leading (past and present) enterprise SSD
companies including Fusion-io, Seagate, Toshiba and Virident. I had already
written my introductory piece above (about how credible technical literature is
a necessity for marketing new enterprise products) before I looked at Principled
Technologies and saw that this is precisiely the service which they deliver.
hidden marketing segments for enterprise flash,
test systems news stories,
play it again
Sam (as time goes by) - the problem with flash write IOPS,
SSD performance characteristics and limitations
looking back at many years of gigs with the data recovery market
September 14, 2018 - Recovering data from damaged storage media (magnetic,
optical or semiconductor memory) in the absence of a usable backup is one of
those subjects which intersects with many technology disciplines.
- reliability - oems can learn about their design weaknesses by engaging
with real world failed drives
- government agencies - need to recover unique data from deliberately
or accidentally destroyed storage
- consumers - may have precious photos or documents on a drive which was
never backep up
- enterprise users - may discover that a single mode failure such as
sysadmin error, new software install or site-wide calamity has trashed their
data and backups too
year at about this time when the hurricanes hit the US the data recovery pages
get a spike of readers - even though they are rarely updated. It's one of those
things which triggers mixed emotions. I'm sorry that anyone needs to look this
type of stuff up. But I'm glad if anyone finds that the articles empower them
in their onward recovery journey.
- financial companies - may need to trace deliberately altered server records
In my new home page blog on StorageSearch.com -
thoughts about data recovery - I look back at what I learned about this
market - where a simple transaction with a credit card can propel you straight
into high tech spook technology. ...read the
Burlywood announces Series A funding
September 12, 2018 - Burlywood
- which is developing software-defined flash controller architecture for
hyperscale data centers - today
it has completed its Series A funding with proceeds totaling $10.6 million.
See also:- a
history of VCs in SSDs,
SSDs in the cloud
SNIA enters the computational storage market
September 12, 2018 - In its
matters newsletter this month SNIA said it is forming a
new work group looking at
SNIA says - "SSDs and other storage
devices now have compute capabilities. How do such devices inform the host and
other clients, in a standard way, of their capabilities? How does the host
program these devices in a vendor neutral way? Those are some of the questions
the new Computational Storage TWG will tackle."
comments:- as a departure from its usual way of doing things SNIA says that
during the initial phase of this work companies "do not need to be
a SNIA Member to participate."
Computational storage aka in-situ
SSD processing also associated with
processing in memory etc
was one of the
ideas which changed in 2014 - as noted in my annual summary on
StorageSearch.com at that time.
Thinking about this new standards
discussion and engagement by SNIA I think it's likely that the most powerful
elements of computational storage will be very application specific and led by
cloud customer and proprietary AI industry platform requirements.
means that long before any so called "standards" emerge it may
already be clear who the leading proprietary companies in the market already
are. (As was the case during the early years of
PCIe SSDs) .
in the long term it will be useful to define general software frameworks
- standards - for interrogating and initializing the customizable features of
computational storage products so that a software ecosystem can grow around
doing useful things with a variety of competing intelligent memory systems
having different price points and acceleration capabilities and just enough "compatibility"
to reduce the risks on the road to inevitable mainstream adoption.
also:- storage industry trade
StorageSearch.com - update on the sale
September 3, 2018 - Re the planned
StorageSearch.com (expected to occur at the end of 2018) an
update document has been produced to help buyers with whom I've been having
discussions to self select from a range of options related to the web
properties and assimilation plans and handover services.
A public announcement offering StorageSearch.com for sale was made in June 2018.
not too late to indicate your interest in this.
However - I've still
got a few important articles I'd like to write before I'm finished and start my
retirement on December 25, 2018.
Megabyte went through
I like the idea that "fastest" claims are still commenting on Violin Systems - October 3, 2018
hard currency in
Septembers of yore in storage market history|
|September 1998 - StorageSearch.com was
still in stealth mode and I was getting ready for the launch in October. What
was going through my mind at the time? I couldn't remember exactly so I looked
in my emails at that time.
what I found.|
September 1999 -Western Digital
the recall of 400,000 recently shipped desktop hard drives due to a
manufacturing defect found in chips supplied to WDC in the month before the
announcement which their QA assessment indicated could result in the hard
drives failing to power-up after 6 to 12 months of full-time use.
later years - in an SSD context - I learned that WD invested enormous resources
into verifying all the component design aspects of storage drives including how
many times they could survive being powered up and down.
In 2009 a few
quarters after WD had acquired SiliconSystems Gary Drossel,
VP of Product Planning, in WD's SSD business unit emphasized how big was the
investment made for long term testing. He joked that the large number of their
SSDs now undergoing long term tests in WD's labs would have almost made the Test
Labs one of SiliconSystems' top 10 customers not so long before.
cultivation and nurturing of "reliability" in a 2.5" SSD brand
September 2001 - We'll never forget the shock to the world
of the terrorist attacks on 9/11. But in the week before that there was
the usual trickle of regular storage news from which I've picked this.
by Adtron was a 6U VMEbus form factor SCSI-2 (that's
storage blade which could be offered with either 5.4 GBytes of flash SSD storage
or 120GBytes of hard drive storage.
You'll note here that the HDD
version has capacity which is 22x bigger than the SSD.
those days (2001) Adtron was making some of the highest density SSDs in the
world so you can take that comparison as a limiting ratio for the 2
technologies in a COTS context.
In 2018 (which is when I'm writing
this) those comparative ratios
are well on their way to switching to the other way round.
highest density flash storage can provide 7x as much uncompressed data capacity
as the highest density HDD in the same physical space. And once you take
compression into account (along with usable performance) the SSD to HDD
maximum usable data storage densities have changed places.
September 2002 - InfiniCon Systems
the general availability of the first commercially available I/O system that
employed the new InfiniBand networking standard... marking the advent of new
data center deployment architectures.
In that press release InfiniCon
said - "Leading analyst firms have projected that approximately 50 percent
of servers shipped by 2005 will employ InfiniBand as a high-speed networking
Editor's comments:- As we know now that
didn't happen. But it did seem plausible at the time because earlier (in June
2001) a news story from InfiniSwitch said "Research from (a well
known company) estimates that more than 75 percent of all servers shipped
in 2004 will be shipped with InfiniBand connectivity."
partly for those reasons that in a news comment (July 25, 2007) I first used
the term "Storage SoothSayers" to describe this type of vendor
leverage of market data in the early stages of technology markets. And I later
added "SoothSayer" as a semi permanent heads up alert in the
market research companies
list on StorageSearch.com in
SSD market predictions and interpretations in those days
were even more prone to include high degrees of uncertainty due to the
disruptive nature of the SSD market - and even when I was writing such
That's why in 2013 I penned the cautionary
blog - Can
you trust SSD market data?
Going back to InfiniBand.
was a combination of competing alternative technologies (the increasing
number of processor cores, and faster ethernet) which meant that InfiniBand
never got near its original dreamed for market share.
highlights on StorageSearch.com - 2000 to 2018.
equivalent of early InfiniBand is Gen-Z.
But you can never be too confident where markets will go. NVMe fabric is
getting better too.