 |
.... |
What were the big SSD ideas of 2015? by
Zsolt Kerekes,
editor - StorageSearch.com
- December 15, 2015
See also:- SSD news in 2015,
SSD history,
RATIOs in SSD architecture |
|
What were the big SSD ideas of 2015? The
new ideas to assimilate in your thinking, and any big ideas to unlearn and
forget. And is it possible to predict yet the SSD ideas which will dominate
our strategic SSD thinking in 2016?
Having been in the professional
SSD market prediction
game a long time - I thought I'd give these questions my best shot.
Unlike
last
year my 2015/16 list (here below) is shorter.
(Big sigh of
relief from those of you out there who tell me they read nearly every word...)
(Similar
sigh of relief from me too - as there is less to write. But also feeling some
pressure that the smaller list better be good. Anxious thought - did I forget
something?)
What were the 3 new big SSD ideas of 2015 to learn?
big
idea #1 - No SSD company is too big to be acquired.
There were
enough examples of big SSD companies being
acquired in
2015 already
at the time of writing this to make the above statement seem almost too obvious
- so why did I add it to my big ideas list?
It's because when the
implications of this idea sink in - it will change the way that vendors and
users in the SSD ecosystem behave - compared to what they might have done
before.
I think the effect will be to favor some specific
changes in technical design and architecture adoption which otherwise might not
have happened at all, or would have had smaller market revenue impacts.
Here
are some immediately obvious changes that I'd expect to see as a result.
In
the SSD vendor ecosystem - it is no longer safe for companies to assume that
their existing technology buying and selling relationships will remain stable
and secure.
Although there was always an element of such risks in
earlier years - for example when
SSD software companies
were being acquired as if they were going out of fashion - or when SSD drive or
array companies which appeared in the
Top SSD Companies Lists
were likely to get picked up by anyone with a spare half billion dollars who
wanted to get into the market - these transactions took place against a
background assumption that there would always continue to be a flow of other
similar companies in the market (if you really wanted that kind of product)
whereas the framework assumptions about which companies would be the biggest
suppliers at the top or bottom of the food chain (as memory makers, or drive
makers or systems companies) mostly remained the same as these T-Rex companies
were comfortable in their dominant market defined identities .
Vendor
strategies which once appeared safe - such as
- this is where my company fits into the market ecosystem, and
- that's where my raw suppliers fit in the market (compared to my place)
and
- that's where my customers fit in the market, and
- this is a company I would choose to partner with
are no longer
safe assumptions when your biggest supplier can become your biggest competitor.
Or when your biggest customer can become your competitor.
And all
these changes can happen in the space of time it takes to read the latest
acquisition press release.
This will lead to more SSD companies
choosing defensive design strategies to decouple their exposure to sudden
changes with the kinds of companies which traditionally they would have
considered as natural partners.
A defensive technology strategy in
this context means acquiring or developing deeper technology IP while at the
same time leaning more on de-facto market standards.
A defensive
business strategy could be to deepen end-user oriented focus and
specialization. In this line of thinking small to medium size SSD companies may
decide it's less risky to be a leader in a small niche than to be a generic
supplier in a big market where the traditional customers have become vertical
integrators and don't need you to exist at all.
The implications for
users of the "No SSD company is too big to be acquired" idea come
from the alternative translation that "none of my big suppliers is safe
anymore - just because it's big."
Your favorite products and
product features may disappear forever in any quarter.
Your safe old
big supplier wouldn't have made the legacy product obsolete - because they
relied on customers like you to keep them fed. But their new owners will have
to be ruthless in their support of the technology they've acquired. For their
own plans to succeed they will have to prioritize the one or two things they
liked in your old supplier's goody bag. Sorry your platform isn't in the list.
I
explored in more detail the defensive changes which enterprise users can make
in their buying and technology adoption behavior in my article about
enterprise
consolidation. In other markets like the embedded industrial and military -
the assumption that your favorite big supplier may exit your market at any
moment - is simply business as usual. (And it has been that way ever since
digital chips replaced the analog, pneumatic and clickety clackety stuff which
came before.)
big idea #2 - there's no single best place to locate
all the IO and management intelligence of a big SSD.
Another way
to say this is that the optimum design solution is to place a little piece of
intelligence everywhere it can make enough of a measurable difference (to the
total application context in which it operates).
When I listed "adaptive
intelligence flow symmetry" as one of the 11 key design symmetries in my
2012 classic article -
how fast can your
SSD run backwards? - it was located pretty close to the end of the article.
That didn't signify its relative importance in SSD architecture. But
it was a topic whose significance at that time was relatively little
appreciated outside a small group of designers of enterprise PCIe SSDs and
related software. And there were very few publicly known examples of product
implementations of this technique which I could point to in my narrative.
In
2014 there were more examples which flashed across the SSD news screens of those
with the alertness to notice such things. And to make sure that my readers got
the alerts they needed I noted this technique as key idea #1 in my
12 key
SSD ideas which changed in 2014 article. Although in that case - it was a
much narrower interpretation of the benefits of this technique - which was
speed. Whereas - like all powerful SSD design techniques - the general concept
can impact a wider range of SSD attributes such as power consumption,
reliability and cost too.
In case you're still unsure what I'm talking
about and how this relates to what we saw happening in 2015 - some of the
reasons for this unclarity are:-
- the technique started in PCIe SSDs, then was next adopted in arrays of
SATA SSDs but these implementations were mostly hidden as "improvers"
in proprietary products.
- the technique was called different things by its various different
creators. These different implementations also differed in the set of
parameters they were intended to optimize. (But actually they were all
variations of the same basic architecture idea. The big idea #2 above.)
Still
no wiser?
In 2015 there were enough new COTS implementations of big
idea #2 to swing it for me as a done deal
See what you think with these
company and word associations...
Fusion-io,
Memblaze (3rd
generation), NxGn (in-situ
SSD processing), Seagate
(cloud customized SandForce),
InnoDisk (FlexiRemap),
Skyera, Baidu (SDF
for web-scale), Radian
Memory (cooperative flash management ),
OCZ (host managed SSD
technology)...
Now you've got it.
Without it - all SSD
application architectures look like they're missing part of their oxygen feed
to the brain.
You're going to see a lot more in the next 2 years - for
the obvious competitive reasons.
big idea #3 - retiring and
retiering enterprise DRAM
which includes a new value proposition for
enterprise flash SSDs (flash as RAM) and presages a rebalancing of server
memories - DRAM will shrink as a percentage of the physical RAM - which will
also make it easier for emerging alternative memory types to be adopted by
hardware architects and by systems software too
In 2015 there
were significant and tangible product announcements around the ideas of
rethinking
enterprise RAM architecture (2014) - which have the impact to change the
balance of memory types used in enterprise systems in as fundamental a way as
SSDs themselves were predicted to change the server, storage and software
markets in my 2005 article -
5 User Value
Propositions for buying SSDs.
You could say that not the least
impact of big idea #3 will be to add a 6th value proposition to the original 5
which I had in my list. And the economic impact of this - could be as great as
any of those earlier 5 ideas.
The size of the business opportunities
represented by retiring and retiering DRAM (aka storage class memories) have
become apparent by analyzing the gains made possible by earlier generations of
enterprise SSDs - in particular PCIe SSDs with their related support software.
If you want to see more detail about this take a look at
what's RAM really? and
DIMM wars - the
Memory1 episode.
Companies which made product or technology
announcements in the big idea #3 context during 2015 included:-
The
above companies were joined in 2016 by:-
Xitore. (Others will be
added to this list as they emerge).
What were the big SSD ideas
which 2015 taught us to unlearn, discard and forget?
Are there
any big SSD ideas or ways of thinking about the SSD market - which were useful
in the past - but whose relevance and usefulness has been downgraded by
developments in the market?
In previous years I suggested these:-
- 2012
- I said enterprise users and systems integrators should stop worrying about
the type of flash in their systems. "Good design is more important than
good memory."
- 2013
- I said that the apparent traction and staying power in the Top 10 section
of the Top SSD Companies
List researched and published by StorageSearch.com would end and that we
would see new names breaking into the list.
- 2014 - I said
we could forget old ideas about the limits of server memory due to new SSD
based software and the business confidence of the SSD ecosysytem that it no
longer had to be tied by backwards compatibility with historically set limits
and zoning ideas in enterprise architecture.
So... what's new to
forget?
It arises as a direct consequence of the way that some vendors
will be implementing the new big SSD idea #3 above and is related to
retiring and retiering RAM. It's this...
It's no longer universally
true that flash is always used as non volatile memory. Instead flash memory has
a potentially viable latent duality of market roles as either:- nvm or vm.
This
is because nand flash has now discovered a new market role which has the
potential to displace tens of billions of dollars of DRAM in new application
roles (as big data DRAM replacement) in which the intensive R/W and
flash care schemes are
not focused on preserving the remanence of the memory.
Sure - if you
can still offer a flash based DRAM replacement that is persistent (like an
SSD) that's a product you can sell. But if the cost of maintaining such good
data persistence compromizes the performance you can offer in that emulated
DRAM - then maybe in some applications it's a worthwhile trade off.
I've
touched on the enormity of this design rethink in some earlier references this
year in my endurance
and DIMM wars
Memory1 articles.
What it means for
flash controller
designers is a break away from traditional thinking about flash care schemes
which was guided by the pre-eminent need to preserve the "n" in the "nvm"
description of flash chip arrays.
Once you say "we don't care
about the n in nvm" that opens up an entire alternative universe of
timing and R/W management which we presently don't know much about. But I'm sure
that as more vendors enter that market space - we'll see hints and details
emerging in papers and patents.
So the next time you see a text book or
blog which describes the primary and distinctive characteristic of flash as
being "non volatile memory" - you'll know it's out of date.
What
are my new predictions for the dynamics of the SSD market in 2016?
I
think that hovering in the background and influencing many seemingly
incomprehensible product changes in 2016 will be the apparently conflicting
tensions between 2 primary Higgs boson-like SSD forces:-
- an urge towards greater standardization, and at the same time
These
tensions will affect all the main markets in which SSDs are used:- enterprise,
consumer, industrial, military, medical etc - although the resulting outcomes
(in the shape of product and systems design) may be hard to recognize at first
as being the result of these 2 tensions.
The urge towards greater
customization will be driven by the need to improve the
efficiency of SSDs
(cost of raw materials and competitiveness) and also technical characteristics
(performance, power consumption, reliability etc) which are optimized
specifically for well defined application specific needs. So whereas in earlier
phases of the SSD market - the viability of new application segments could be
prototyped and proven using mainly simple variants of COTS SSDs - it is
imperative from both the user and vendor competitiveness points of view - that
SSD products are designed to be a better fit for the purposes in which they are
intended to be used. In that respect any intrinsic features which are not
recognized or valued by the applications segment become instead frivolous
luxuries which simply add to the cost and detract from other opportunities to
provide systems apparent benefits.
The urge towards greater
standardization will be driven by the needs of users (and vendors) to have some
degree of continuity in applications and business roadmaps against a backdrop of
more chaotic than usual technical changes at the raw memory and (massively
changing) architecture assumptions.
Software will play a big role in
both these primary forces.
And a factor which will make the
analysis and interpretation of some new SSD product lines harder for external
market commentators is that we'll see both customization and standardization
forces being implemented simultaneously in the same products - but in different
parts of the design.
Here are some specific examples of how the above
2 forces will interact to drastically change aspects of the SSD market which we
have hitherto come to regard as being mature and well established.
customization
and standardization - the SSD directional push-me pull-yous
- enterprise systems
Customization pressures - means we're
going to see more efficient
products (lowest hardware production cost needed to do the job) being designed
for each type of clearly defined product
role. An inevitable
consequence of this will be a demise in business prospects for sloppily defined
"general purpose" systems which straddle application boundaries and
which are therefore not the most competitive solutions for any application. This
missmtach in most cases in the market today stems from bad
marketing
thinking rather than technical design.
Standardization pressures
- will be good for the business of any software vendors who credibly offer
standard SSD software platforms which aren't tied to any specific storage
hardware vendor. (While at the same time the software vendors can demostrate
they have a sustainable business model which isn't simply to get acquired.)
How
many different hardware platforms do you need in the support list to establish
a "standard"?
A rough guide would be as few as any 6
hardware platforms (of vendors who have ever appeared in the Top SSD Companies
List). That would give users the comfort of knowing that even if half the
existing hardware platform vendors disappeared from the market - they (as
users) would still have a range of competitive suppliers among those who
remained.
I discussed these issues in more detail in an earlier article
- 90%
of enterprise SSD companies won't survive - here's why
- SSD controllers
Customization pressures will mean that
there will be many more application uses for SSDs which cannot be viably
satisfied by using standard COTS
SSD controllers. This
means that more systems companies will resort to designing their own controllers
at the raw processor
level.
Standardization expectations - have meant that we've already
seen many newcomers to the SSD controller market who think they would like to
take a percentage of what they see as a growing pie.
- industrial SSD market
I noted the growing degree of
customization business strategies being observed among
industrial SSD
companies (last year) as one of the
big SSD
ideas of 2014 - which I summarized like this...
"Designers
have refocused and chosen the viable reality of excellence in selected niches
above the unfe asible goal of having the best technology roadmap for all
applications."
Now, looking ahead to 2016 - I think we'll see
more of those marketing intentions being visible in new products. A key sign
will be more SSDs which use more heavily customized, and proprietary SSD
controllers rather than lightly tweaked standard controllers (similar to other
markets).
Special additional arguments for doing this are the
demanding industrial constraints imposed on these designs, such as :-
environmental factors (temperature, vibration, humidity, physical space etc),
power consumption (and robustness) and (unlike in the enterprise) the greater
reliance in industrial systems on the intrinsic
reliability of
standalone SSDs (rather than arrays).
Standardization
influences on the industrial market will come from many directions.
- commercialization (by means of COTS controller support) of new emerging
high temperature alt nvms (which previously had prohibitively high integration
costs).
.
- market pull from the clarification and definition of new distinct IoT
roles for SSDs.
.
- adoption and transfer of mobile phone reference platforms (and OSes)
into industrial equipment applications.
.
- industrialization of flash as DRAM (enabled by migration of software from
enterprise markets).
.
- growing strength for the argument that using standard FPGAs (where
feasible) as the controller implementation technology - assures customers of
longer BOM availability
than COTS processors.
That's it from me for now. It's
fascinating to think that the excitement about changes in the SSD market and
coming from the SSD market will continue to be at the forefront of digital
systems thinking and architecture in
2016.
See
also:- the
limericks of flash SSD endurance | | |
.. |
 |
.. |
StorageSearch.com
/ popular SSD
articles /
SSD market
history |
| |
.... |
 |
the annual SSD market roundup |
big idea #3 - retiring and retiering enterprise DRAM
which
includes a new value proposition for enterprise flash SSDs (flash as RAM) and
presages a rebalancing of server memories - DRAM will shrink as a percentage of
the physical RAM - which will also make it easier for emerging alternative
memory types to be adopted by hardware architects and by systems software too
|
.... |
 |
.. |
RAM
news ain't what it used to be
After AFAs? -
cloud adapted memory systems
memory
controllernomics - an interview with Diablo
4
companies whose SSD technology showed the way ahead |
.. |
the SSD
consequences of DRAM latency creep |
We've got used to the fact the performance and
characteristics of
fast flash SSDs
owe more to the interface, controller
design and software
than to the intrinsic timing and reliability of a
nand cell. But what about
DRAM? Isn't that much
closer to the raw, organic memory ingredients?
In a new blog on
StorageSearch.com -
DRAM's
indeterminate latencies and the virtual memory slider mix - I look at the
growing gap between server memory performance and the capabilities of raw,
organic DRAM cells.
As enterprise RAM performance is progressively
bounded more by controller traffic analysis needs and power consumption rather
than DRAM cell capability - there's an argument for saying if the statistical
distribution of DRAM latencies is so wide - maybe no one would notice if you
slipped in a different kind of memory in the virtual slider mix.
And
that - of course - was precisely the open barn door - which ushered in the
explosion of DRAM lookalike alt nvm announcements in
2015. ...read the new
article | | |
.. |
|
.. |
 |
.. |
I was talking to an end user whose organization has spent hundreds of
millions of dollars on EMC storage.
They'd love to decouple
themselves and benefit from modern lower cost flash.
But the flash
marketers in startups aren't doing those kinds of conversations.
For
many of them a single customer like that is bigger than their whole business
plan.
|
"compared
to EMC" - the unreal positioning of AFA startups | | |
.... |
industrial SSDs in 2016?
my prediction - "industrialization
of flash as DRAM (enabled by migration of software from enterprise markets)"
was
confirmed in May
2016 - when Marvell
showed how its Final-Level Cache technology could replace DRAM in portable
scale products like phones.
|
.... |
 |
.... |
memories?
We used to be able to write down a list of those which
mattered.
Now we're seeing many new strangers from strange lands.
Even I can't remember who they all are and how to spell their names.
the SSD
Bookmarks |
.... |
|
.... |
After 2003 the only
technology which could displace an SSD from its market role was another SSD
product. |
SSD market
history | | |
.. |
After a particularly bad day - reading contradictory SSD
white papers, blogs and market reports you may be inclined to ask
yourself...
Is it time to update my profile on linkedin?
Can you
trust SSD market data?
|
.. |
other
key SSD ideas in 2015?
top SSD companies...
|
In the notes to the left of
this I've focused on significant market wide SSD trends rather than significant
SSD companies.
For a summary of 2015 as seen from an SSD company
list perspective - see these articles below:-
| | |
.. |
|
.. |
 |
.. |
|
.. |
|
.. |
|
.. |
One of the things which
demonstrates the extraordinary range of diversity in thinking about the SSD
market is the different answers you get to these 2 simple questions.
what's the best way to design a flash SSD?
where's the best place to put it?
Nowhere else in computer architecture will you get so many industry
experts disagreeing on such fundamental questions. |
the SSD Heresies | | |
.. |
|
.. |
|
.. |
What do want our memory
systems to do for us?
And how do we want them to behave? |
can memory do
more? | | |
.. |
"We looked at where
the large corporates failed to satisfy their clients, such as when low profits
and inconvenience caused those big companies to be disinterested, we then
approached the clients with customized solutions"
|
Kevin Wang, VP
Sales - Longsys
- talking about business growth tactics in the blog -
Longsys taps into
IoT market with custom-made solutions (June 14, 2016) | | |
.. |
 |
..
..
.. |
"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." |
all enterprise data
will touch an SSD | | |
..
..
..
.. |
|
..
..
..
.. |
|
.. |
now consider this...
90% of the enterprise SSD companies which you know have no good
reasons to survive.
The main reasons they will hang around a little longer and their ranks
may even swell slightly before the big shrink are bad reasons...
Bad marketing and an inefficient market.
routes
to consolidation in the enterprise |
.. |
| |