(And how did we get into this mess?)|
|by Zsolt Kerekes,
- which explores the notion - that if there were equal
amounts of matter and antimatter in the universe shortly after the
- then how is
it that we're not seeing more antimatter everywhere now?|
I like a
good mystery. But I'm still stuck in the early chapters. Who would have
believed physicists could make up so many bit part characters? Don't email me
to tell me who stole it - or how it all ends.
For many years the
SSD software industry
was like that too. And it would be reasonable to pose the question...
the SSD market is so big and important - then why haven't we seen more SSD
I've been thinking about that topic for a long
time. And in the early 2000s when something called
was the fashion in storage software - I asked some of these new HSM companies
- where's the SSD layer in your standard model?
It seemed to be
missing from all their pyramidy diagrams.
It soon became clear it
was missing from their thinking too.
rotation speeds of hard drives - they could manage those OK in their storage
hierarchies - but SSDs weren't part of the picture.
I thought - it's
amazing how ignorant storage software companies are about SSDs. But they weren't
alone in that respect.
Then a few years later I thought maybe the
operating system would be a good place to locate the SSD software.
In 2004 I
company called Sun Microsystems they could put SSD support in their Solaris
OS to make their servers appear faster. They didn't do it.
And a few
years after that when Microsoft started talking about putting support for
balancing the load between HDDs and SSDs in Vista it seemed like something
useful might come from that direction. But it didn't. Later Microsoft sold off
its SSD caching technology and it wasn't good enough to make any of its
subsequent owners rich.
It was still a mystery to me how we ended up
in a situation - where throughout most of the
history - there was precious little software apart from drivers and some
I would date the start of the true SSD software market
at around 2009.
That's when we started to see shipments of products which promised
auto-caching and auto-tiering - which I lump together as
SSD ASAPs. But
there's a lot more to SSD software than that.
With the benefit of
hindsight and having talked to many founders of leading SSD software
companies I think there were several reasons the software industry seemed to
take so long to wake up to the idea of SSDs.
- the SSD market was too small and fragmented.
If you're selling
software software at somewhere between $100 to $2,000 per license - you need a
big established hardware market to fund a software company.
- everyone thought that someone else was going to do it.
think that SSD management is going to appear in a future release of the big
operating systems - then it's not worth investing the effort yourself. Instead
you wait to see what the OS companies do - and then hope to create products to
fill in the gaps.
But the big OS companies didn't do anything. They
were used to getting technology roadmaps from the processor makers - which
told them exactly how much memory and what type of interfaces they would see
in a 5 to 10 year lookahead timeframe. And SSDs weren't in those plans.
makers too discovered out they weren't in control of this new SSD market
either. The best they could hope for was to
buy in SSDs from
the market until they could figure out any better plans.
several years - it seemed like the most productive thing that traditional
storage software developers ever did for SSDs was to search for and remove
countless embedded delay loops from their legacy code which had originally been
inserted by performance tuning geniuses (who had since moved on to higher
management) to prevent thrashing HDD controllers with too many I/O requests.
- SSDs were more complicated than other types of computer hardware - and a
fast moving target.
It's clear from conversations I've had with
software developers that many people invested time in understanding
complicated aspects of SSDs which later turned out to be irrelevant or short
It turns out they were mostly chasing phantom problems like
block size performance sensitivities and endurance. Good SSDs would deal with
those problems internally.
They should have been thinking at a
different architectural level and instead of just asking - how can we make SSDs
fit into our old HDD software schemes? - asking better questions like - what
can we do with SSDs that we couldn't do before? Or - what new opportunities
will SSDs everywhere create? (Some did ask that question.)
the SSD software market is making up for lost time now.
And in the
past year or so - a good marker of that progress has been - by how many SSD
software companies have been acquired. The acquirers have been almost always
been SSD hardware companies.
Why would they do that?
business answer - is that if the software makes it easier to use SSDs - or
easier to use more SSDs - then making it compatible with your own hardware is a
better investment than recruiting more sales people or spending more on
For SSD hardware makers - leveraging software around
their products is probably the highest added value activity they can do.
is why we've seen small software companies gobbled up for amounts of money which
just don't make sense when you look at how little they would be worth valued on
We're now in an exciting time for SSD software. In the
next few years we're all going to hear about companies we've never heard of
before - doing stuff - some of which (already) looks predictable - but some of
which is entirely new.
What's the role of systems software?
In the past it has always been about optimizing whatever was seen as the most
At various times and in various places that "most
valuable resource" has meant different things:-
and managing the
data was mostly
- feed the processor,
- make many processors work as one,
- make them talk to each other,
- show us some pretty pictures,
- manage this unruly population of hard drives.
- backup the data (tonight please)
- stop the bad guys getting to the data (how come those malware guys are so
smart at writing software?)
the SSD era the most valuable part of the infrastructure is the SSDs.
- unback the data - (who would have thought it would be so difficult).
SSDs systems designers have the freedom to ask fundamental questions like - if
data is so valuable - how can I automatically create even more valuable
data and then ensure that it's in the right place at the right time to do
good for my business?
selling data can be my business?
Good SSD software will change the
rules about everything you can do with SSDs.
Where will you find out
There's always been coverage of SSD software - here on
- it's just that for many years there wasn't much substance to
The interplays between the
memory chips, the
interfaces and applications
are poised to branch out in many new directions.
There are many
questions still remaining to be answered.
- where's the best place to put the SSD software?
In the OS? In
the cloud? In the app? In the SSD controller? Or a little bit everywhere?
- What are the boundaries of where the SSD software should reach?
it have boundaries? Or should it - like the internet - reach everywhere?
- If the SSD software is so important - then should we be more careful about
whose software we buy and what features we use?
now... no-one knows the answers.
- Will SSD software stabilize into being a commodity - where everyone's
software does nearly the same? - or will it be like databases and OS's - where
you started out using it because it was convenient - but then 20 years down the
road you're locked into a solution that costs a fortune to migrate .
But the key people who will create
those answers - or choose which casts of the SSD software dice will succeed in
the market - like you - also read the pages of .
|| And when the time is right - we'll all find
out together. Stay tuned to this channel.|
|re - Your
article on "Where are we now with SSD software?"|
I just read your article today on "Where are we now with SSD
software?". I found it to be a very interesting read.
In particular, you have touched upon several important items, such as
reasons why the software industry has taken so long to wake up to the idea of
SSDs - and moreover, the "many questions still remaining to be answered"
that you listed at the end of your article.
Indeed, it does appear that we are entering into an exciting time for
SSD software. Thanks again for being at the forefront on this important topic.
Best, Tom West
hyperI/O (May 29, 2012)