| leading the way to the
new storage frontier|
|"You may decide that
my ratios are too timid" |
I said to Skyera's CEO - Rado Danilak
so - scare us!"
by Zsolt Kerekes,
editor - October 22, 2013
As I discussed in an earlier
the rackmount SSD
market is an exciting hot spot for developments right now.
not talking here about IPOs,
although they're exciting enough for stakeholders in the hot seats or those
hoping to hold the hot tickets - but I'm talking here about real technical
Rackmount SSD enhanced storage is exciting today because
there are now things you can do with software and hardware architecture at the
complete storage array level - which are outside and beyond the reach and ken
of the cleverest controllers
within any individual SSD drive.
I hinted at the consequences we can
expect to see when you factor in all the architecture and software changes which
I know are in the pipe (both within server-side and SAN SSD storage) in my
new SSD software
event horizon blog. (Perhaps an alternative title might have included the
phrase "inflexion-point" - but anyway it's out there.) I've had some
good reactions to that article from people inside the industry - although like
many of these articles it may be a year or so before these factors are shaking
the SSD market so much that the concepts become widely known.
was curious to get the reaction of one particular company - which for several
reasons which I've described before - seems to me to be at the leading edge of
more of these technology trends at the same time in a single product line than
any other single company (which is not in stealth mode). So I set up a meeting
You probably know who they are - because they were one of the very
few companies in SSD market history which in the same quarter as they emerged
from stealth mode also made it into the Top SSD Companies List. That was back
my email last week to the founder and CEO Rado Danilak
I talked about my event-horizon article and invited him to talk to me about
these concepts. This is a cut and paste of what I said....
My key point
is that with modern designs and new architecture and new software - it doesn't
take a petabyte of SSD
to replace a legacy petabyte of HDD at the enterprise or cloud level.
The kind of ratio which can be achieved in the next few years is very
much higher. So high - in fact - when you take into account the possibility of
apps and operating systems which will be rewritten for SSD - that people had
better get used to the idea. That's it really.
Obviously the ratio of what is do-able now will improve - as the SSD
software ecosystem gets more confidence.
You may decide that my ratios are too timid. If so - scare us.
And that's what led to an hour of me talking about the enterprise
flash storage market to both Rado Danilak (CEO) and
Tony Barbagallo (VP
Marketing) at Skyera.
Early in the conversation I said that I and
many competitors in the market I've discussed this with - regard Skyera as being
the company to beat in terms of
cost per terabyte
of rackmount SSD
on the SAN - above a floor
capacity level - and within some defined compatibility segments.
I also know that due to the cautious nature of the enterprise market - and with
Skyera being a new company with new technology - I wouldn't expect them to get
much real business traction within their ideal customer group - until after the
obligatory 2 years waiting period on hardware reliability and software
stability which faces all newcomers to the enterprise.
you had all your perfectly baked enterprise SSD cake - with all the software
trimmings which are still a year or so in the future - available right now in
your restaurant and offered it to these people at that futuristically low price
today - they wouldn't want to taste it" - I said. "They don't want
to risk being poisoned by Rado's crappy flash SSD juice. They prefer other
people to experiment with the new e-SSD chef in town. You just have to live with
that. In another few years these same cautious types may be saying to everyone
they know - don't you just love dining at Skyera?"
One of the
difficulties is that people don't really understand enough to appreciate what
Skyera has got and is offering. It looks to me - I said - that you could
thrash around with blogs and interviews which skirt around these design issues
- but most people still won't get it. Others - the most experienced SSD users
would understand. They're the ones who matter. (But understanding what you do
and being interested isn't the same as buying the product in huge numbers yet.
(Although some dark
matter users might.)
That got me onto the idea of suggesting a new
channel partners program for the upcoming skyEagle (which gives you a petabyte
of SSD in 2U) to educate
the market about the positioning of this efficiently designed flash array.
My suggestions were along the lines...
Tony liked the idea - and I said he could use it in future ads or
communications if he thought it would work. (If you work at another SSD company
and want me to suggest a better way to explain what you do - your know where to
- Wrap it in a bigger box.
- Add bricks to make it heavy.
- Add a heater to make it run warmer.
- Double the price to make it more acceptable.
We got pretty quickly into - how complicated it is for
people to understand what's really important in the enterprise SSD market.
I've been in the enterprise market for a long time and my experience
goes something like this....
Every year I learn 2 new important new
ideas about SSDs.
But every year I also have to remember to forget
or discard 1 old idea which was vital to know before because it's no
longer useful, valid or true.
"What made you go into the
enterprise market?" - I asked. This was particularly aimed at Rado - whose
previous products have been used wherever SSDs are used - but which wasn't
particularly focused on enterprise systems.
Rado said - "I had the
benefit of the brick wall of ignorance. Not knowing what couldn't be done."
went on to say how once he decided to optimize around the flash, that meant
throwing way conventional interfaces, that mean he needed more software to tie
it all together.
As I've previously elaborated in my article on
controller architecture - there are optimizations you can do when you have
access to a bigger arrays of flash - which are impossible when you're trying to
optimize around the single drive level - or arrays of drives level.
said he had an initial list of efficiencies which he thought he might be able to
do - but he said the more you work with large amounts of flash in a system the
more you understand - and he's now in a position where he and his team will
revisit older assumptions they had - which gives them the ability to expand
their efficiencies by impressive multiples - even taking where they already are
the starting point.
Rado told me that the really impressive efficiency
numbers (in flash utilization) which are being worked on - are all in the
software. (That's similar to what other array vendors have said too.)
said that in Skyera - it's already the case that 3/4 of the engineers are
working on software - and Skyera is hiring more software people - so that ratio
Rado asked me about the raw assumptions I had used in doing
the scaling and efficiency projections in my article. Mostly I said - they had
come from earlier articles (and I sent him the links) but collectively the
calculation of what I thought was do-able was simply based on combining all the
different bits of IP that SSD companies tell me about - churning them through
some simple architecture sanity tests - and then applying some discounts and
backtracking for the things which might not work as well as had been originally
I said (in a later follow up) that whatever crazy numbers
I come up with in some long range SSD forecasting article - the market usually
beats that number by 1 or 2 years - or by a factor of x2 - due to something
innovative being done which was previously unknown or thought to be impossible.
(I may add the detailed assumptions for readers as an appendix later.)
Unlike the designers and software developers who have to make these all these
impossible things work - all I have to do is steal the top level idea of what
they're doing, assume it does work and then model what effect that has on a
complete SSD installation. Because the sum of the parts includes the app servers
- and SSD ecosystem-aware software. It's about more than just scaling up
efficiencies in the storage arrays. The results are impressive.
spoke about segmentation
within the enterprise and why the old fashioned ways of describing customers,
products and apps use cases don't provide adequate models in the SSD case.
the SSD context it's frequently the case that different customers in the same
market, or the same customer within different parts of their own organization -
will choose solutions which appear
different at the raw technology level - simply because some products are a
better match for their requirements for compatibility or ease of managing risk
and roadmaps at that point in time.
The same customer - with 5 years
more experience of using SSDs under their belt may start new projects with a
completely different approach to how they want to see things moving along - and
the level of incrementation they prefer to work with - at the upgrade or
level. No single product can satisfy all these conflicting needs.
wondered if I should quote more of what Rado said in this report about the
kind of improvements he says are still to come in his own company's roadmap.
I said I think it would be best for me to hold off for now.
(and the enterprise SSD industry) need to do a better job of explaining the
consequences of some of the remarkable things they've already done (or can do)
with their new systems. When that "do-able now" is firmly grasped -
then the "do-able next thing" will seem even more impressive (or
I promise you will get more updates on this topic soon.
inside the SSD box
internecine SSD competitive advantage
meet Ken and the
enterprise SSD software event horizon
SSD endurance -
the forever war
history - since the market began