enterprise SSDs - exploring the
limits of the market in your head|
|This article looks at the usefulness of
analyzing pivotal market boundary conditions and stress tests to help you
understand what's going on now and what could happen in the future enterprise
SSD market (without needing a PhD in electronics).|
by Zsolt Kerekes,
editor - September 2012
|A critical test of whether you really
understand the dynamics of a complex market like enterprise SSDs - is whether
you can predict what rational buyers might do when offered new product options
at the extreme limits of - for example - price.|
That's why - in the
thousands of conversations I'd had about the SSD market - I've always been just
as interested to learn about users doing something which the vendors hadn't
expected - but which worked out OK in the end - as I have been to learn about
management techniques and
this raw material goes into the SSD market model grinder - although it can be
many years before it's clear whether what was predicted - turns out to happen
When thinking about SSD market boundary conditions the
starting point is often... this is what we expect most people to do - if we
make the usual set of common sense assumptions.
But what if we
change some of the assumptions? Maybe stretch them to breaking point. Maybe
change what we're comparing things to? Is there a point where the market would
behave in a completely different way? And what can we learn from that?
mental modeling tricks and stress tests can change markets - for those who
believe in them - because they get can reveal insights about the future by
using a cheaply available tool - the human imagination.
When I talk to SSD companies - an interesting part of the conversation
is often trying to figure out how products which barely exist yet will compete
and fit into an infrastructure which doesn't yet exist anywhere and which itself
may rely on several dependent SSD market revolutions which haven't happened
Resolving the gaps and equilibrium points between different types
of SSD products and technologies happens in these discussions not only many
years before they hit the market - but as part of the thought process for many
SSD company founders who can see the gaps for themselves and have good ideas of
how they will fill them with a new startup business venture.
past - most of the SSD predictions that people were interested in were related
to SSD's impact on the hard
I used boundary condition analysis and user value
propositions as the launchpad for several influential articles which described
SSD inroads which went way beyond what the conventional analyst wisdom was
saying at the time to predict what turned out to be useful market roadmaps for
a bunch of SSD vendors. These early examples were:-
- My 2003 article - could
SSDs become a $10 Billion Market? - which indicated that the enterprise
SSD accelerator market wouldn't remain a niche market of last resort for
performance distressed users - but would instead grow to become the mainstream
multi-billion dollar market we're seeing emerge today.
also correctly predicted BTW that once any one of the main server companies -
HP, IBM, Dell or (at that time Sun) - started shipping factory installed SSD
accelerators - all the others would rapidly have to follow suit to remain
My analysis there showed that it was irrelevant how much
SSD capacity cost compared to HDDs. The only relevant factor was how the cost
compared to the alternative of more server CPUs - which I later named SSD-CPU
The conclusion was:- SSDs could make servers run faster.
SSDs meant users could do the same workload with a fraction of the servers. Or
they could run their apps faster and still use about 50% less servers.
2012 - year of the
enterprise SSD goldrush - this is now the conventional wisdom. A
survey (August 2012) of visitors to VMworld suggested that over 60% of
attendees already have SSDs in their datacenters
- My 2005 article -
5 User Value
Propositions for buying SSDs
- explored all the forseeable markets for SSDs at that time.
was clear that even within the price competitive
market - there were use cases in which SSD based notebooks would be attractive
for some users even when SSDs cost high multiples of HDDs.
market growth was initially delayed for many years due to badly designed
notebooks which were mostly old HDD designs with SSDs installed as an
Where are we now?
- My 2010 article - this
way to the petabyte SSD - was stuffed with many technology ideas - but
as several SSD vendors said to me in the months and years following its
appearance - the key message they came away with - was that SSDs will replace
HDDs as raw storage capacity devices in datacenters - even if the hard drives
That was the key foundation of the article.
else in it was exploring the technology and market ecosystem which would lead
to that market boundary event - the end of HDDs in the enterprise.
It should be clear from the above
notes that most of my early applications of testing SSD market boundaries were
driven from the perspective of seeing how enterprise SSDs would fare in their
role as the memory barbarians at the datacenter gate. In those days the SSDs
were on the outside battering their way in.
In the past few years I've
turned the focus of my analysis to try and understand how different types of
SSDs will compete with each other.
SSD market boundary analysis can be
used to answer a whole bunch of interesting (and sometimes less interesting)
questions - which I pose from time to time as the backdrop for new articles -
or more often - are the reaction to reader questions such as:-
- how will vendor X be affected by new products from vendor Y?
- what's the future of server side caching vs caching on the SAN?
- will PCIe SSDs make SAS SSDs obsolete?
- is there a future for this that or the other interface / memory technology
- or new product idea etc?
Even though I didn't explicitly say so at the time
(March 2012) my
SSD Silos article -
which lists the sustainably different types of SSDs that can co-exist in the all
solid state storage datacenter of the near future - came into being after a
series of sanity tests in which I inserted these boundary concepts into
real time conversations with many SSD industry leaders (and also email
dialogs with selected readers) over a period of several quarters to see how
well they stood up.
- what's the role of SSD software? - and which type of product or company
The result? - was to add more granularity to the
different SSD types - as I realized the market would be big enough to support
more distinctly different speed classes of SSD rack than I had envisaged in
earlier versions of my model just a few years before.
I love it when I get anecdotal feedback from the market
in stories which provide some more meat on the flesh of these bare bone
concepts and shows that the outline predictions were correct - even if the
exact factors determining user preferences were unknown (or unknowable) at the
For example - in my 2011 article - written as a response to
reader questions about PCIe SSD vs rackmount SAN SSDs (if Fusion-io sells more
does that mean Violin and others will sell less?) - I concluded that it
was a mistake to think of this as an Either / Or case - even if in some
applications (such as a first time installation) either would solve the problem.
I confidently said that...
"PCIe SSDs make the locally attached
server faster. But SAN SSDs (including Infiniband) will always be needed for
environments (legacy command and control business architectures) where data is
shared across more than a threshold number of servers. Another complication is
that a PCIe SSD itself can be a component inside a rack such as bunch of cards
rack (Nextio) or as an internal cache (NexGen)."
to the present - in August 2012 I heard 2 examples confirming this analysis
from "competing" leading SSD market vendors.
Director, Marketing Management at
SanDisk told me
about a customer they had in the financial services market who knew they could
meet their performance goals equally well from the technical point of view
with either a server-side acceleration system (using PCIe SSDs) or a bunch of
rackmount SAN SSDs.
Petersen told me that in this case - the SAN
solution would have resulted in a lower hardware cost. But instead the customer
went for the more expensive option of adding more PCIe SSDs into their servers.
"This customer had the money to afford whatever option they
wanted" he said. "And they had previous experience with SSDs..."
said the reason they went for the server-side acceleratuion was that all their
SAN based storage infrastructure was tied into security, reliability and legally
mandated data management processes which would have added a ton of extra
support costs to the SAN solution. Making their servers faster was easier -
because they were managed in a different way.
Meanwhile - in another
conversation at about the same time which confirmed another customer's
preference in exactly the opposite direction - Thomas Isakovich,
CEO of Nimbus Data
Systems told me about a customer they had - who was already using a
significant bunch of PCIe SSDs (from
Fusion-io - and had a
lot of experience with SSDs) in their servers - who had decided to buy high
availability rackmount SAN based based SSDs from Nimbus to solve their
performance problems - instead of more server side SSD accelerators.
may be saying more about that in a future case study - so I won't go into
details here - but the inference you can safely draw is that if a user is on
the roadmap to multiple petabytes of SSD and has an organization that stretches
across more than a few thousand feet (the present limits of cheap PCIe
clustering) and across multiple sites, and if the user has legacy style apps
then the need for SAN based SSD isn't going to go away.
predicted the sustainable co-existance of both types of SSDs in exactly the
same customer sites should come as no surprise.
There are many other
examples I could have given here.
market boundary analysis - in which various assumptions about technology, price
and markets are stressed to breaking limits can yield useful insights into
possible future market behavior and help to determine the positioning of
product classes and vendors upto several years before these trends become
clearly observed in the market.
This type of thinking can also be used
to test the viability and probable market success of blue sky SSD products
which don't yet exist - but which are being explored to fill market gaps
identified by other types of analysis.
This technique can be applied to
problems ranging from the trivial upto problems which are market wide in their
It's fun to do. I enjoy it. And the outputs from this type of
thinking have yielded useful results in the past for those - like me -
who have been looking for clear trackways through the coagulating chaos of
the SSD market present of today through to a clearer picture of the possible
Once you get into the habit - it's hard to look at the SSD
future any other way. You can become your own
Although I hope you'll still come back to these pages to compare notes from time
what else is there?
Buyer behavior is only
one aspect of understanding SSD market futures.
Another useful stress
testing analysis technique I've found is exploring gaps and application fits
in SSD technology and - in particular- imperfections and asymmetries in
SSD design architectures.
For those of you who are more interested
in technology than marketing babble - I've already written an article on that
very subject. It's called
how fast can your
SSD run backwards?
Here are some more articles in the "SSD thought leadership"
series - which have appeared on the home page of StorageSearch.com
|It looks like you're seriously interested
in SSDs. If you've got the time - you might want to take a look at
the top 100 most
popular SSD articles here on the SSDmouse site.|
|If someone could offer a
memory system which had the same storage density as mainstream RAM - but was
infinitely faster - could we use it? - how much would that be worth? and how
would it change markets?|
|are we ready for
infinitely faster RAM?|
|As you know from market
history - the proposition that one new SSD thing can replace one old SSD thing
is rarely as simple as the advocates of the new thing say. |
|PCIe SSDs versus
memory channel SSDs |