12 key SSD ideas which changed in 2014by
editor - - December 18, 2014
|key SSD idea #1|
This is about closing important gaps in the
intelligence of message passing and the speed of data access between
application processors and SSD controllers.
(SSD controllers which -
in the vast majority of SSDs - come attached with their own offload processors
or associated data movement engines.)
Traditionally SSDs have been
designed to reduce the access times to data - but within the framework of
commands, APIs and data structures which have been designed for applications
agnostic data storage drives.
And in order to make SSDs easier to
use - classic SSD controllers also perform a lot of house-keeping and
related functions - in a way which is the apps processor doesn't need to know
about (although it can collect stats related to endurance etc).
is much evidence to support the idea that both applications performance and
data storage efficiency
can be greatly improved if the flash management and apps management processor
are either the same CPU - or if they can talk to each other in a more
These improvements (which enable data handling
responsibilities to be partitioned downwards to the flash or upwards to the
apps host - depending on which has the best view of what is needed) have
already been implemented by various SSD vendors in experimental or custom
These concepts - which first took root in large web scale server
farms - have also delivered useful results in some industrial SSDs when
scaled down to a single SSD.
A handful of SSD vendors have already
done work in this area - notably
Fusion-io (in whose
non offloaded SSD
controller architecture - the flash management and apps processors were
the same processor).
2015 and 2016 these
concepts will become much more explicitly talked about.
SSD array companies (who probably already do some degree of in-situ SSD
- 80% of fast enterprise data IO is transient - but legacy storage software
often sucks this up as part of its reliability functions and then attaches a
similar weight of metadata and overhead as if it were archive data. This is some
of what I learned in various conversations with David Flynn
cofounder Primary Data
|key SSD idea #2|
This year - as part of a continuing trend - we've seen an
upswing in the number of companies who offer
PCIe compatible SSDs
in form factors like M.2 and 2.5".
The barriers to market have
been reduced by standards such as NVMe and SATA express - which by creating
frameworks of software and hardware interchangeability - have minimized the
risks for oems who incoprorate such SSDs into their storage and computing
An important new factor for the PCIe SSD market this year
was the materialization of product announcements centered around the core
concept of using PCIe as an interconnection fabric between racks.
key pioneers driving these efforts have been
|key SSD idea #3|
random access memory doesn't have to be RAM
The idea of using
flash as a new memory tier isn't new. And neither is the idea of using flash in
DRAM memory slots. But in 2014 there were several developments which added
weight to the usefulness of these ideas.
- Applicable to any kind of standard flash SSD -
software (described by as "one of the most significant
SSD software products launched in 2014") is an API toolset which gives
software designers the freedom to treat flash in a similar way to DRAM - thereby
being able to rely on much higher capacities within any given monetary budget
Although the performance characteristics of such memory
won't suit all applications - the ability to experiment and invest in a
technology platform which promises to avoid lock-in to any particular SSD form
factor - will encourage the development of new types of data repurposing
are you ready to
rethink enterprise DRAM architecture?
- Those who may have been disappointed by the low aspirations of
Diablo's 1st generation
memory channel SSDs - were given a glimpse of something more akin to what they
might have been wishing for - in the unveiling of an ambitious 2nd
generation architecture which promised to go much further in 2015.
key ingredient here is a new software framework (Carbon2) with features like
The new software is being offered as part of
developer packages which anticipate 2nd generation MCS hardware which will be
fast flash DIMMs compatible with DDR4.
|key SSD idea #4|
micro tiering and micro clouds
One of the trends in computer
architecture in recent years is that new software architectural concepts which
deliver sustainable efficiency or management efficiencies have found it easier
to get their benefits established and recognized at a large scale - as part
of big web entities or cloud infrastructure.
But the lessons learned
have been duly noted and reapplied to other use cases and are now finding
their way into individual rack scale products too.
3 companies which
stand out for their different approaches in this respect are:-
|key SSD idea #5|
R/W (including DSP) data integrity management in flash
2 years ago
- there were only 10 companies with
technologies in their SSD product lines.
It was important to know
who they were at the time.
Because looking ahead from the perspective
of 2012 they and their licensees or acquirers were going to be among the first
vendors who could leverage the economics of next generation flash.
did this by moving away from classical flash controller technologies - which
relied on anonymous industry wide characterization statistics for key flash
parameters - and moving towards an adaptive model - which was able to
recognize and grade different qualities of individual flash blocks (even
within the same SSD).
The new adaptive DSP technology was able to
choose from a wide bandolero of timing and ECC techniques instead of being
dependent on a single caliber flash manage bullet.
By the middle of
2014 - adaptive R/W had become a mainstream technology - deployed by most
leading enterprise SSD systems (in applicable products) - so its strategic
advantage as a competitive differentiator has diminished.
has become the new "standard technology" for handling all sub 20nm
planar MLC flash devices.
But it would be wrong to think of it as a
uniform technology. There are significant differences in the scope,
granularity and associated controller and power footprints of the many
different adaptive DSP flash IP sets used in the SSD market.
|key SSD idea #6|
nand flash -may be tough enough for industrial markets
nand flash SSDs have been shipping in the market - the current technology
doesn't deliver enough efficiency and cost advantages to replace 2D in the
short term. Many manufacturability and design problems remain to be solved
before that is likely to happen in mainstream SSD markets.
On the other
hand the raw
1st generation 3D flash seems to be 3x to 4x better than 2D at
the same line geometries - according to early work done by an
company FMJ Storage.
these early impressions are confirmed in later volume production - this could
open up the possibility of alternative markets for this type of flash.
also:- flash memory news
and articles, DWPD -
endurance in industry leading enterprise SSDs
|key SSD idea #7|
reported in 2014 seemed to indicate that SSD companies aren't worth as much as
they were before.
Although there are special factors which complicate
any particular analysis - as I discussed in the cases of
LSI's SSD business,
and SanDisk acquiring
Fusion-io - it's
clear that from the viewpoint of the people who matter (those with the money)
an SSD company with a rich set of IP and strong market recognition in 2014
isn't generally worth as much as you might have thought if you had extrapolated
from SSD company values in 2013.
Why is that?
In one way it
seems perverse - given that the overall market opportunity for SSDs is now
generally assumed to be much larger than it was before.
I think the
key factor at work here is evidence (as reported in financial reports of some
leading SSD companies) that competition is much tougher than before (due to
the growing number of competitors and also the rise in the quality of such
But another key risk factor (for any encumbent SSD
vendor) is vulnerability to future technology shocks - which can disrupt
their business prospects.
These technology shocks don't just stem
from new startup SSD companies - but can also occur as a result of macro changes
in the market as users
change the way they use and deploy the same type of SSDs when using
|key SSD idea #8|
pricing and business models
How much should you pay for an
enterprise SSD array?
And what exactly is it that you're getting?
SSD vendors had always been enthusiastic about what their products and
technologies could do in the first decade of enterprise flash - the language
with which they bundled their pricing offers did not show the same leaps of
creative imagination which they were expecting their customers to make.
in 2014 - a small number of SSD pricing pioneers designed new enticing
pricing models for their flagship flash arrays which broke away from the
formulas of the past.
Behind these new pricing models was the explicit
recognition that there is always a high degree of uncertainty involved in
such purchases for various technical and business reasons.
This was the
subject of my previous 无限彩票app下载 page blog -
the Astrological Age of Enterprise SSD Pricing - which describes why the
change is happening now and names the companies who are leading this charge.
|key SSD idea #9|
Surprisingly - given its already substantial size
and gravitational business pull for SSD drive makers - there are still
significant parts of the enterprise SSD market which remain uncharted and
For investors and SSD startups the opportunities to grow
business in under exploited high value user territories may be a source of
comfort - given the potential upside.
However, for users who are still
waiting for vendors to offer them the kind of products and services they
really need - it's a source of frustration.
I described the reasons
for these market voids in a recent article -
hidden segments in the enterprise.
In some embedded markets -
which use a lot of SSDs - the rackmount SSD is simply viewed as a dumb component
- like a 2.5" drive. It's a component with larger capacity and more
performance - but is simply one of many. And although it's a box - it's not "the
system". In fact its internal cleverness and associated software are
sometimes regarded as a nuisance by true applications aware systems software
which has a much better idea of what's going on then the little SSD box
designers could ever have imagined.
|key SSD idea #10|
importance of SSD software
One of the key ideas which permeates
everything now in the SSD market is the importance of software to the SSD
an article in
January this year I said
"the SSD software market is getting
ready for a world in which all enterprise data touches SSDs"
elsewhere in the same article I also asserted
"the winners in
SSD software could be as important for data infrastructure as Microsoft was for
PCs, or Oracle was for databases, or Google was for search."
from any confirmatory events in 2014 - when I think about to what might
happen in the next few years - the overwhelming importance of SSD-centric
software seems like a no brainer.
I think we will see not only new
predictable generations of SSD software coming to market (which will be
designed to work with currently known computer architecture models) but also
entirely new data architectures and ecosystems whose very existence has
been predicated on the assumption of a widely deployed SSD enhanced base
|key SSD idea #11|
industrial SSDs - designers have refocused and chosen the viable reality of
excellence in selected niches above the unfeasible goal of having the best
technology roadmap for all applications
I have talked to many
companies this year - and there is definitely a different mood in the air
about this market and some confidence that vendors can carve sustainable
business niches - having found (differing) rational strategies to cope with the
chaotic changes in the general SSD market.
Those companies which still
have industrial SSDs as their main product lines - when many other companies
have exited this market - due to the siren pull of bigger markets (such as the
call to consumer SSDs in 2006, and the big pull towards enterprise flash
which was hard to resist by 2008) have survived several waves of turbulent
change in their own market in the past 5 years.
At the start of that
period - in 2009 - 2011 - the first wave was due to the impact of
commoditization in the SSD market due to the success of the
merchant controller market
- and in particular SandForce
In 2011 to 2012 - the long held assumption that MLC
flash would never be good enough for industrial applications was called into
question by the apparent disproof of that very notion in many enterprise
products. That caused customers to question - why can't we use MLC in
The practical questions for industrial SSD designers
- looking at MLC were:-
This was a difficult balancing act for
industrial business owners - because even if they solved the problem of
sourcing and managing
reliable enough MLC for some applications - with new controllers - the added
cost of other factors in the design - due to the increased hold up time needed
to clean up block management operations in MLC compared to SLC (new firmware)
- and the lower capacities used in many industrial systems - meant that the cost
benefits of making the transition to MLC - were not always clear cut or
Customers told me they experienced a distinct lag in the
market of about 2 years - during the transition in the top 20 or so industrial
suppliers from talking about the availability of MLC in their products -
while others were actually doing it.
Part of that was due to
competitive market differences (some companies do things faster than others) but
another factor was a fundamental difference in views about whether that was the
right solution for all products. (It still isn't.)
What has become
clear in 2014 - is that there is now a greater degree of specialization
within the industrial SSD market.
This has come about because no
single company has a single set of IP which is most competitive for all form
factors and interfaces.
- when it comes to controllers - industrial SSD makers have different
approaches even within their own product lines.
The diveristy of
industrial controller solutions goes beyond the simple filter of performance /
form factor / power consumption and memory type - and standard versus in-house
design - to encompass firmware adaptions of standard controllers, and stretches
to customized firmware which can optimize system performance for known
configuartions and software environments.
And industrial SSD companies are also finding new
markets in the enterprise too - in hot spots in blades and small solo SSDs
which are used in managing services rather than as primary storage.
- The industrial market represents a bigger total available market than ever
But set against that is the need for greater specialization -
and application specific optimizations.
The result is greater market
fragmentation - and more niches - rather than a small set of big broadly
aspects of SSD customization
|key SSD idea #12|
SSD designers will adopt any kind of naughty flash - once they've figured out
what to do with it - and have validated the memory in less intense consumer
10 year trend
- in 2014 - 3D MLC indisputably joined the roster of flash types deemed good
enough to ship in enterprise SSDs - notably confirmed - if there were any
doubts - by the announcement in
September 2014 -
that Samsung was using
3D nand in a new PCIe SSD - rated at 10
DWPD for 5 years.
Currently there is no type of mainstream nand flash which isn't
being used in some type of enterprise SSD systems.
And if you hear
vendors say - that their array is better because it uses so called enterprise
MLC (eMLC) it really means that they don't know how to manage the flash with
their own IP and have passed the buck to their
memory suppliers and to
their customers (who have to pay more).
In some high end enterprise
market applications - there are valid reasons you might choose to
pay more for your
flash and have your flash array delivered in a bigger box - but in most
applications - that choice is a
Maybe you like the software which comes with the box -
or it will cost you more to validate alternative suppliers. But eMLC is not -
and has not been for many years - a necessity in most enterprise flash arrays.
the other hand - if you are a worrier - rest assured that the reliability of 3D
nand will need to be reassessed in future generations as the stack layers
progress upwards in number. (Bad things might still happen.)
|23 years later... and still
counting - as in 12 key SSD ideas, Top 10 SSD Companies etc|
Where does - 23 years later come in?
Earlier this month
(December 2014) LinkedIn's software picked up the fact that it was the 23 year
anniversary of my having founded the enterprise
Although 23 is a prime number - and in
that way is - I suppose - interesting for some - it wasn't a milestone I had
planned to write about or mention on these pages. But as some of you picked up
the bot generated posting and said nice kind things I thought that deserved
some kind of human generated response from me. So here's what I said...
for your kind and motivational comments. In the new era of SSD guides the
balance of effort has moved away from - what are the new products and
technologies? - To inferring - where they are - on known and unknowable
intersecting roadmaps of evolutionary and disruptive change - with destinations
which don't yet have words in the jargon of computer architecture."
publishing and social networks
Everyone has their own preferences. In
my case - I've become accustomed to my "social informative network"
being my readers rather than any of the new fangled channels such as LinkedIn
or Twitter. So this web site is where I invest most of my efforts.
it's not always obvious what I've been working on - and a reader asked my
about that recently. He said he hadn't seen anything from me in recent months
about the SSD market - and asked if everything was OK.
In his case it
was simply that he had recently joined a Top 10 SSD 无限彩票app下载 - whose corporate
servers deny its employees access to - because their firewall
filters out sites which have cartoon-like content. I won't name the company -
they've had this internal problem for over a year. I guess my readership would
be bigger otherwise. The marketers in that company like the mice and they like
the content BTW.
After our email exchange - we spoke for about an hour
about strategic changes in the enterprise SSD market. Very interesting - but
for background - rather than a new article.
Apart from corporate
firewall walls which block out from some desktops - another
reason you might not see what I've been writing about recently is that I know
you won't be interested in it yet - so I don't flag the links loudly.
I choose to write about a new SSD topic - it's because it's interesting or
important to me - and my gut tells me that it will be of interest to seriously
minded readers like you at some time in the future. Sometimes it can take 5 to
10 years for new SSD related ideas to get into the mainstream market. But
by writing about them early - I'm able to begin a conversation with the blue sky
architects and business visionaries who will do the difficult part - which is
making these things happen - and fit into a world which wasn't created for their
The truth is - there isn't enough time to complete half the SSD
articles I get started on. Too many interesting things going on.
time to time - it's useful to distill the essence of all that raw random SSD
chatter into something simpler.
That's where the above article - 12
keys ideas etc - comes from.
|other key SSD ideas in
top SSD companies...
In the notes above I've
focused on significant market wide SSD trends rather than significant SSD
For a summary of 2014 as seen from an SSD company list
perspective - see these articles below:-
the Top SSD Companies
SSD endurance -
trust SSD market data?
how fast can your SSD
the top 50 SSD
meet Ken - and the
enterprise SSD software event horizon
SSD market consolidation - why? how? when?