Booting up a dialog about the future of
PCIe SSDs
Editor:- July 24, 2012 - Texas Memory Systems
has joined the bootable enterprise PCIe SSD set. This feature is now available
in all new RamSan-70
models - and is also available as a firmware upgrade.
Editor's
comments:- Bootability in PCIe SSDs is a useful thing to have for some
types of applications although it's not a new idea. Companies like
OCZ - whose PCIe SSDs are
arrays of small
architecture controllers have offered bootability for years leveraging
2.5" SSD
technology.
TMS wrote a whole
press release
about this - because they thought it was a big deal being the first big
controller architecture SSD companies to offer it.
But when I spoke
to Erik Eyberg at TMS
about this yesterday - I said while I could see that it might make a
difference to TMS's business having this feature - because it would open up new
design slots - I didn't think that many other editors would be interested - and
maybe only about 10 readers.
I explained that I expect that
all PCIe SSDs will have bootability in a few years - as part of adapting to the
needs of the 2.5"
PCIe SSD market. So I said - can we talk about something else
instead?
I ask nearly everyone who designs SSD hardware what they're
doing about adaptive
DSP IP in SSDs. So I started with that - although it mainly affects a
segment of the enterprise SSD market -
fast-enough SSDs -
which TMS hasn't been traditionally associated with.
TMS has been
generally conservative about its use of flash and Erik said that while they're
talking to a lot of memory companies about their roadmaps TMS SSDs mostly use
older 32nm geometries - therefore the DSP techniques aren't yet a must-have
item.
I suggested that some memory makers will simply remarket leading
edge flash but with "enterprise" R/W characteristics evolved from DSP
IP - so they could charge a higher price - as with
eMLC. And
that increases the designer's range of options even more.
I also asked
Erik if TMS had looked at the 2.5" PCIe SSD market yet - because having
bootability meant there was no impediment to doing that (provided they could
squeeze in enough memory chips around their fat controllers in that confined
space to make it worthwhile.)
Erik said TMS has experimented with
smaller SSD form factors in their labs - but it's not on their immediate
product horizon.
Erik asked me what I thought would happen to the PCIe
SSD market? - Did I think it would eventually be dominated by the semiconductor
companies?
That was a good question...
I said I had thought
that might happen - when I was first asked this question about 4 years ago -
but the PCIe SSD market didn't go that way - and now that the picture of what a
PCIe SSD has to do is much clearer - I considered it even less likely that the
semi companies would dominate - although they might get lucky and score well in
a couple of big oem slots and be among the leading vendors in 1 or 2 memory
generations.
I said there's too much software and complicated systems
engineering needed to design a good enterprise SSD.
The guys who run
memory companies already have difficult jobs. They have to understand business
cycles, big investments and memory roadmaps. That's difficult enough to fit in
one brain. Until SSDs are the biggest outlet for their memories - they won't
manage the SSD businesses as well as pure play SSD companies - who aren't tied
to fabs.
Also - I said - chipmakers have historically been really bad
at software. Even if they buy an
SSD hardware IP
company or a bunch of SSD
software companies - after 9 months to 2 years - the hot talent will get
fed up and leave.
I've said before (on these pages and in conversation
with these companies) that the big semi companies may have to buy maybe 5 or
6 SSD IP companies as part of their SSD learning experience - and even then -
if their own memory is late to market - then a competitor who isn't tied to one
bunch of fabs can deliver a better SSD to the market which is cheaper to make.
And another thing is - even if SSDs rate high enough attention in
the memory company - who's going to call the shots? The consumer or enterprise
groups? They have very different needs to make them work.
I said to
Erik - that in another (human) generation - when you have SSD people running
memory companies - that's when they might start to dominate the PCIe SSD market.
Until then - and for the forseeable future - it will be a diverse set
of companies with different slants which satisfy different segments within the
PCIe SSD market or whatever repalces it. (That's going to be the theme of a
future article BTW - SSD actors in search of a play.)
Erik and I
also talked about the importance of low latency and some of the various ways to
do it. He said the experience of TMS is that it's very rare among the customers
they talk to - that they would be willing to risk the lock-in of using new APIs
just to get more performance.
I said it's not surprising - because on
the legacy vs new
dynasty divide - TMS's products are legacy. But I think the market for
green field new solid state storage businesses and apps will make the new
dynasty market just as big - and maybe in the long term bigger too than legacy
storage.
In the meantime vendors which offer SSDs for each of these
markets don't really compete head to head - even if the performance of their
products looks similar.
And that's how a 45 second conversation about
bootable PCIe SSDs grew into 45 minutes of something else. | |
| . |
|
| |
....... |
|
