There are a couple of bookstores in my neighborhood. They could not be more different from one another.
The first is known to focus on best seller lists, to promote popular books, and use displays and traditional retail techniques to drive business. They seem to do well, year in, year out. The other bookstore is more of a community treasure, beloved by the neighborhood, with a focus on the (thoughtful) insights of their staff. Those insights are delivered via small note cards appended to shelving throughout the store, where books are displayed alphabetically, with library-like neutrality.
The first store is very market focused, changes with the season, and seems to be quite succesful. The latter store, beloved though it may be, struggles to stay in business.
Now you might consider that an awkward introduction to a discussion of datacenter storage devices, but it was the best I could muster after reading yet another week’s worth of horrible economic headlines, dominated by multi-billion dollar losses and bailouts for some of the world’s largest corporations. But I (really) don’t want to talk about the economy (or the irony of a Wall Street analyst with a sell rating on Sun asking me for a job). Nor do I want to talk about the economics of selling books. I do, however, want to talk about storage and flash memory, far more scintillating topics (to me, anyways).
As you’re aware, flash memory is making its way into data centers at a very rapid clip. Flash has exceptional performance characteristics, and is generally orders of magnitude faster than a traditional disk drive at responding to requests to read and write information – up to 100’s of times faster. It also requires little/no power, dissipates no heat, and can withstand vibration, temperature extremes and shock. Plus it’s got that great name, “flash memory” – who wouldn’t want that in their datacenter ( (myths about flash being less reliable than traditional disks debunked here).
Sun recently introduced a storage device heavily optimized with flash memory. It’s known by a similarly scintillating name, the Sun Storage 7000, and is what we call an “open storage” device – one built with commodity flash and disk components, alongside open source software (this is an excellent summary).
The 7000 has one remarkably interesting attribute: it learns. The longer it’s doing its job, interacting with applications and serving data, the faster it becomes. How it accomplishes this relates to the bookstores I first discussed.
Most storage devices behave like the unsuccessful bookstore, organizing books with a sorting algorithm to which the user ascribes no value – disks make no attempt to position data based on the frequency of access, and as a result, popular content is delivered with the same, or poorer latency as unpopular content. Storage architects go to great lengths to provide band-aids to the problem, but most fail in their attempt and fall back on massive over provisioning. That is, they throw money at the problem by “short stroking” data – making sure all data is written to the outer sectors on a disk platter, the parts that spin fastest, delivering highest performance reading/writing. This can work, but it’s exceptionally inefficient – and wastes money, power, space, disks and patience.
Conversely, the 7000 behaves like the commercially oriented bookstore, and uses algorithms (instead of storage admins) to adaptively place the most frequently accessed data where it will be fastest to retrieve: flash memory. (I was talking to a software analyst today who wanted to know why we bothered with the hardware business when so much value was in our software assets – this is a perfect example, the value isn’t in the software or the hardware, it’s in the systems we build with both.)
The internal operating system (OpenSolaris, paired with ZFS) actually “warms up” the device after it starts working – it watches traffic, notices which files are being most frequently accessed, and caches them in flash. Thereafter, they’re available near instantaneously. The 7000 can then use far slower, and less power consumptive disks for the infrequently accessed data. Net result? Hot files are served up orders of magnitude faster than even short stroked, jet fueled premium enterprise disks (did I mention those disks cost an outrageous fortune?) – and customers spend radically less on the device, on power, cooling, space, etc.
From your end user’s perspective, the front page of your news site, your most frequently accessed products or content, or the Facebook profile of the pilot who just saved the lives of 185 passengers landing an airplane on a river, are served up with lightning speed. Your users are happier, your CFO is happier, your family, with whom you’re spending more time, is happier. (And if there were ever a machine optimized for MySQL, this would be it – but it also runs brilliantly with Oracle, DB2 and SQL Server – and Postgres, of course).
It’s common sense: if you put the bestsellers on the first shelf a visitor sees when they walk in the door, they’re more likely to buy one than if you put them in alphabetical order around the store. As the bestsellers change, so do your promotions and displays – if you adapt to demand, you capture more of it. That’s the basic premise behind the 7000, to use systems innovation to drive performance, eliminate latency and radically cut purchase and operating cost.
That behavior might make a bookstore less beloved in my neighborhood, but it makes Sun more beloved in the datacenter. And it makes the 7000 a great candidate to be one of the storage industry’s best sellers.
If you’re a reseller or customer, and would like to try out a 7000 free of charge, just click here…
There’s also a great simulator available for free download here – so admins can check out the user interface and diagnostic capabilities of the 7000 without installing a machine.
And speaking of best sellers, here’s a video in which the 7000 plays a starring role – showing you how not to coax the best performance out of your storage devices.