|
SSD
companies mentioned in this news period:-
Fusion-io,
KingFast,
Micron,
Nimble Storage,
OWC,
Skyera,
STEC,
Violin Memory,
Virident Systems |
| . |
Skyera launches industry changing enterprise
SSD rack
Editor:- August 14, 2012 - Skyera today
launched its first product - a 1U half-depth 10GbE
SSD rack with 44TB usable capacity ($131,000 approx), 3.6GB/s throughput and
upto 1 million IOPS
using under 800W electrical power.
Integrated software includes inline
dedupe, compression, snapshots, clones and various management functions.
Skyera
will demostrated the new
Skyhawk next week at the
Flash Memory Summit.
Editor's
comments:- last week I spoke to Skyera's founder and CEO Rado Danilak
and some key people in his team - which in addition to core members of the team
who designed some of the early
SandForce controllers -
also includes talented and experienced
people from other well
known chip and enterprise storage companies.
Skyera make a big thing
about their cost value proposition.
They deliver fast SSD storage
at a system list price under $3 / GB (uncompressed) and about
$1 / GB
(with compression and dedupe). That not only beats best of breed SSD systems but
seriously challenges HDD based storage too.
Designed from the ground
up the Skyhawk can be summarized as using
adaptive DSP,
with big
controller architecture (proprietary rather
than an array of COTS open SSDs) and with an integrated
software stack which
screws the efficiency of
reliability and
data integrity
as tight as can be - by intervening at multiple levels (high
adaptive
intelligence flow symmetry.)
In any given year in the SSD market
there are maybe 2 or 3 industry changing announcements - which illustrate what
the shape of the future could be.
Skyera's Skyhawk is one of those
in the year we're already in. And would be just as significant if the launch
had been delayed into next year too.
Here's some more stuff I learned
from talking to Skyera.
Funding
Rado said Skyera has
one strategic investor - which he didn't want to name right now. The company is
mostly owned by its management. That gives them an incentive to design a product
which can make money as soon as possible.
FPGA rather than ASIC -
for Skyera's hardware controller
The FPGA versus ASIC tradeoff
debate in electronic design goes back to the mid 1980s. With FPGA you buy a
chip that's got arrays of logic and common functions already in it. You
define what your FPGA does by programming the interconnections in a way
that's similar to writing to flash. With ASIC you have almost total freedom to
decide what functions your chip does - but the chips are made in a factory
that's making a lot of other chips - not just yours. You're buying time in
someone else's multi-billion dollar factory. There are set up fees, learning
curves and queues involved.
Skyera implements its controller using
FPGAs.
That gives flexbility (also
explained to me as
a factor by Holly Frost at competitor
Texas Memory Systems
- which uses both ASICs and FPGAs in different places - while
Nimbus prefers to use
exclusively its own ASICs and
Fusion-io is still
staying mostly with FPGA.
In a rack system - the recurring unit cost
is not a big deal - whereas it is a
cost disadvantage
in a single small drive - which is why
STEC made much of
switching to ASICs in its enterprise SSDs last year. Where the
VC factor comes back
into the equation is that each design costs hundreds of thousands of dollars -
and months of delay. So if you have the realistic choice of either FPGA or
ASIC in the form factor and power budget - which is not an option if you're
designing a 1 inch
embedded SSD for example - then the FPGA route means you can ship products
sooner with less initial outlay and your revenue stream can contribute to
earning profits faster.
Having said that - my guess is that after
their launch - if Skyera said they needed another $500 million or so to do
something - like remodel the executive car park - VCs would rush in to supply
the cars too. But Skyera probably doesn't need that much money.
Core
SSD technologies
On the SSD architecture side I've already written
about the strategic importance of large architecture, adaptive DSP, SSD
software, tradeoffs
between flash capacity and reliability and performance and the benefits of
integrating them at a high level. So I won't repeat those points here.
Rado
said he wanted to get the best performance and reliability possible out of "cheap
crappy flash". He said if you use a traditional
RAID 6 array - then the
latencies start to mount up - and you end up doing 3 writes cycles to get the
parity protection. That's bad for endurance and bad for latency too. He said
that Skyera's way of doing RAID-like protection means that the write
amplification for a 4kB write is an average of only a factor of x1.04.
That's a 3x reduction on write amplification compared to arrays of
2.5" SSDs for
example. There are many different levels in the hardware and software - where
the self knowledge about the system is used to good advantage.
Their
controller technology can be adapted to work with any flash. In theory they
could populate production systems with the cheapest flash they can get and
even mix up the sources in the same SSD and still get a working system.
Nevertheless the company's deep knowledge of flash - means that (in my words)
there's little risk of cooking the memory cells to get the performance. And they
know which suppliers are better.
Earlier this year
DensBits explained
to me how using
adaptive DSP
meant that you could get 60% faster write performance out of MLC flash
because the DSP gave you the data integrity when using shorter write pulses and
lower energy - which also leads to better
endurance.
(See endurance
multiplication wars.)
Anyway Rado explained - that once they
decided that they had to design a proprietary memory array - they knew they
also had to invest in the software layers to make it work within the enterprise
infrastructure.
Internally their box is like an array of memory
connected to a switch. In that respect - it's similar to the RamSan systems from
TMS.
High
availability
Skyera's initial system isn't a
true high
availability SSD - and although it has many essential enterprise features -
this will limit the appeal of the product in some applications. Nevertheless, I
know that many super users of SSDs would prefer to use their own cloud like
wraparound software - and all they need is a low cost scalable rack which they
can use as a component to build multi-petabyte SSDs.
Skyera said they
see traditional enterprise SSD apps like analytics being a good fit for the
launch system.
HA is in the product roadmap and has hooks in the system
architecture. It will come later.
Business development
I
suggested to Skyera - that given how
popular their
company had been with SSD readers while they were still in stealth mode - the
problem they might have after this product launch would be too many inquiries.
That would place a burden on marketing to filter out the type of SSD leads
which were a good fit and those which were just curiosity.
Maybe we can
recycle the leads we don't want and sell them to other SSD companies - said
Skyera.
The SSD market will be watching Skyera with interest to see
if stacking all these new technologies together for the first time in a
system really works.
Later:- when WDC acquired Skyera they
stopped supplying the Skyhawk racks which meant customers had to go elsewhere.
But that didn't prove too difficult as the enterprise rack market was already
warming up for the marathon of
user
independence and vendor consolidation.
OWC has new SSD upgrade for Mac market
Editor:-
August 14, 2012 - OWC
announced that it has a
480GB
SSD upgrade option ($579.99 MSRP) for the MacBook Pro with Retina display -
for users who need more capacity than the factory installed 256GB SSD limit.
SSD cliff coming to memory makers sooner than to HDD oems -
that's why they're tired of waiting
Editor:- August 14, 2012 -
Outside of SSDs there is no viable long term future for mass memory makers.
Years ago when I first started writing about that concept - what I had
in mind was that SSDs would eventually become a more important market for high
volume memory makers than supplying components as boot devices and traditional
computer motherboard memory - because the memory capacity populating SSDs
would become orders of magnitude higher than anything which the memory makers
had ever seen before in traditional design slots. Also the profit - if there was
to be any - would be inside the application dependent SSD - rather than outside
the SSD party looking in as a merchant memory supplier to SSD companies.
In
those days the SSD market was about 100x smaller in revenue than the
memory market - so I knew the market transformation would take time - and it
would be easier for smaller memory companies to transform into SSD focused
companies sooner.
What has only become clear in recent years - is that
there's another acceleration factor at work in the memory to SSD business
transformation model.
It's that next generation small geometry
flash memories (of the
type which could compete in mass markets) don't work
reliably as
standalone components and in order to be used by anyone they have to be
wrapped around with some kind of complex
SSD-like controller
technology. Even if you don't care about whether
hard drives have a
future or not - the SSD cliff is coming to semicos sooner than that. If you're a
volume memory chipmaker - you don't have a future without SSDs in a time window
which is quarters rather than years.
Investors get nervous when they
read about that kind of thing - and particularly if they discover that the
memory market has to re-engineer itself into some kind of solid state storage -
much sooner than the message they were getting from the hard drive companies.
That's the reason you're starting to see SSDs starting to get
prominent mentions in the financial presentations done by memory companies.
One recent example - from Micron -
Micron's
Flash-based Storage Vision (pdf) - tells you how they see the SSD market and
their position in it.
Don't get too attached to the exact numbers in
this document. The presentation under-sizes the
enterprise market in
my view. But it shows that the SSD market is getting serious attention at a
higher level and in wider circles than ever before.
See also:-
Hostage to the
fortunes of SSD
Violin plugs some software gaps with Symantec
Editor:-
August 13, 2012 - Violin
Memory today
announced its
rackmount SSDs can now support snapshots, cloning, dedupe, replication and
thin provisioning - based on software IP from Symantec. See
also:- where
are we now with SSD software?
FIO's results and various SSD market success factors
Editor:-
August 10, 2012 -
Fusion-io's
revenue - in the recent quarter (ended June 30) - grew 49% year on
year to pass $106 million, while revenue in FIO's fiscal year was $359
million (82% up on the year before) - the company
reported
today.
Editor's comments:- a lot of people ask me -
what's the secret of Fusion-io's success?
I've tried to answer that
question from many different angles - and you can see what I've said (on the
web) by
clicking
on this site search link.
My working life in recent years has been
spent exclusively focused on SSD companies.
Talking to them, reading
about them, discussing their technologies, business plans, experiences etc. And,
I have - as you can see in the
SSD Buyers Guide
- offered a variety of lists which you can rationally populate with the names
of SSD companies by analytical measures which everyone would agree with.
But
there's another list too- which I have in my head - which I've never
written down, or published - and which is just based on pure gut feel and
instinct.
I can place almost any enterprise SSD company into 1 of 2
groups.
- those who get the SSD market - and where it's going
You might ask - Why don't I publish it?
Here's
my explanation.
In the
enterprise SSD Goldrush
you don't have to understand where the
enterprise SSD market is
going. You can still get very good business results and satisfied
customers by incremental and tactical advances in products (if you're a
vendor) and projects (if you're a user).
My list of "who gets
the SSD market" - works for me. But it might not be so useful for you.
And
there are better shortcuts which don't involve such subjective judgements
(on which not everyone
would agree anyway) and which don't require any detailed knowledge of what
motivates the thinking inside these companies. Because - and here's a sanity
check - it's always the market which is the final arbiter of success.
That's why (and I can't say this too often) looking at lists like
the Top SSD Companies
is useful. It tells you which companies other people like you (who think
seriously about the SSDs) are giving more of their attention to. I don't rank
these companies. You do.
So it doesn't really matter what I think - at
a gut feel level. Except when it comes to me deciding where I should be
spending more of my time:- such as - which emails should I reply to? which
phone calls should I take? and for how long? and which web pages should I
see? Trying to convert this chaos into a coherent narrative - which might be
useful - is the fun part and it pays for the cheese.
Beware of easy
answers and succcess formulas in the SSD market. Few of them can stand the
test of time. (Even mine.)
With the possible exception of this one.
"In SSDs - rules are made to be broken." ...from the
enterprise
SSD Survivor's Guide
STEC's marketing legacy still imploding revenue
Editor:-
August 8, 2012 -
STEC's revenue
in the recent quarter was halved compared to the year ago period, and is 19%
lower than the immediately preceding quarter - it was
announced
yesterday - and the company has set aside $35 million (approx) as a
contingency to settle claims connected with share activities in 2009."
For
many years I strongly criticized the company's marketing in these pages for its
weaknesses in routes to market and tardiness in extending its technology into
the enterprise PCIe SSD
market.
Recently (June 18) I said I
was pleased to see many signs that the company has made visible changes and
improvements in its marketing. But years of neglect won't be cancelled out in 1
month or 1 quarter - and that's why this quarter's revenue trajectory
still looks so dire in comparison to competitors like
OCZ (54% SSD revenue
growth) and Fusion-io
(anticipated tomorrow - proved to be 49%).
STEC's CEO and Chairman,
Manouch Moshayedi said in the company's release text - ""Less
than a year ago, we began the strategic process to fundamentally shift our
business model from addressing the SSD market solely through traditional OEMs to
a model mixing OEMs with emerging SSD system vendors, enterprises and
non-traditional end-user customers. In order to address the needs of large- to
medium- enterprises, we have been building our sales and marketing
infrastructure to cater to enterprises directly and develop our new vertical
market strategy..."
See also:-
STEC's ziggy zaggy
meanderings in the enterprise SSD market - (2007 to 2012) - a narrative
extracted from commentaries in past editions of the Top SSD Companies.
KingFast's new mSATA 3 notebook and fast erase rugged SSDs
Editor:-
August 8, 2012 - later this month KingFast will be
showing its new SSDs at 2 international events:-
Flash Memory Summit in Santa
Clara, CA and IPC
& Embedded Expo 2012 in Shenzhen, China.
One of the company's
recent products is an
mSATA 3 SSD
designed for for high-end Ultrabooks. While the other new product will be a
rugged 2.5" SATA 3 SSD with
fast erase.
Virident's new FlashMAX II
Editor:- August 7, 2012 -
Virident Systems
today
announced
it will ship a new generation of fast PCIe SSDs in September.
FlashMAX
II (pdf) has upto 2.2TB usable RAID protected MLC capacity, 103K random
R/W IOPS (4kB 70:30 mix), and 1.1 million random read IOPS (512B), and <80µS
random read latency (4kB) in a ½ length, low profile form factor.
Editor's
comments:- I spoke to Shiva Shankar
at Virident about the new product and the
PCIe SSD market.
Virident
sees this type of SSD as heading towards a distinctly different storage tier -
and that's the direction of their software focus - even though in the present
market the products have been designed so they can drop in and work with
legacy storage
software and VMs with minimum fuss.
From the design and marketing
positioning point of view Virident has always placed great emphasis on
application type
symmetry and scalability
symmetry. Shiva told me their performance scales linearly. That means if
you have 8 PCIe slots and install upto 8 of the new FlashMAX modules - the
available application performance will be Nx what you would predict from the
single module results.
Virident also say that their performance doesn't
degrade significantly over the lifetime of the product. They call this "Infrastructure
Predictability" - and say it's on the order of 1%. In contrast - the
performance drop off in some competing enterprise flash SSDs can be more like
20% to 30%. (This is a vulnerability in some flash SSD designs which has also
been mentioned by STEC -
as an argument in favor of their
CellCare technology.
I
asked Shiva if Virident uses
adaptive DSP
ECC techniques in its SSDs - and (as I expected) he said "no"
(because it's a technique which has been deployed mostly to improve the cost of
fast-enough enterprise
SSDs (and components) - whereas Virident is in the fast end of the market
spectrum.
Virident specifically says that its new product is 2x
as fast as a well known competitor. See also:-
the 3 fastest flash
PCIe SSDs
Nimble expands on scalability
Editor:- August 6, 2012
- "Scalability" - in SSDs and storage - means different things to
different people - and this is one of the themes discussed in my
SSD symmetries
article.
So I was curious to see what Nimble Storage meant by
the word "scale" - as it's used more than 30 times in a
recent
press
release from the company about a new range of rackmount SSD ASAPs.
A
blog by Nimble's CEO,
Suresh Vasudevan yesterday gives a more precise explanation of what
they mean - which is in effect - you can scale capacity by adding hard drives or
scale performance by adding SSDs - or do a bit of both - "without ever
disrupting applications."
Nimble's systems span a much narrower
range of performance scalability than you would see across the
enterprise environment
- so I'm not sure if scalability is the word I would have chosen to describe
what they do. "Flexibility and convenient management within a useful
application footprint" is probably more accurate - but doesn't sound so
hot.
Fusion-io does a few new things
Editor:- August 2,
2012 - the performance and strategic importance of
SSD software was
reinforced in 2 recent announcements by Fusion-io.
Yesterday
- FIO
launched
its new ION software
- which is a toolkit for bulding your own network compatible
SSD rack by
adding some Fusion-io SSD cards and their new software to any leading server.
The concept isn't entirely new - because oems have been doing this
with various different brands of
PCIe SSDs for years
and this is a well
established alternative market segment for PCIe SSDs. What is new - is
that it makes the whole thing much easier.
Fusion-io says this new
software product "delivers breakthrough performance over
Fibre Channel,
InfiniBand and
iSCSI using standard
protocols." (1 million random IOPs (4kB), 6GB/s throughput and 60
microseconds latency in a 1U rack.)
Earlier this week FIO
announced
it was collaborating on getting interoperability in server-side flash and
caching software
with NetApp. It's
easier now to write a list of major storage systems oems who aren't doing
something significant with FIO.
Going back to SSD software...
In
the
1990s Sun
Microsystems created and leveraged the phrase -
the
Network is the Computer.
I have long thought an apt
reinterpretation of that in this decade is "the SSD is the computer"
- or maybe the "SSD software is the computer" - because the ultimate
characteristics of fast computers are determined more by the SSD architecture
which is installed - than by the same old CPU chips. | |
SSD
articles
SSD history
SSD market research
|
| . |
|
| |
|
