Anyone paying attention knows I’m not particular sanguine about the near-term prospects for solid-state disks (SSDs) and hybrid hard disk drives (H-HDDs) in the enterprise storage space, but I’m not foolish enough to discount them entirely. With that in mind, it’s worthwhile noting the debut of the first commercially-available retail(ish) SATA SSD from SanDisk. Read more below…
Tom’s Hardware, one of my favorite nests of geeks, tested a pair of these 32 GB SanDisk drives and came out with some interesting findings. Notable among their findings is the fact that the pathetic random write throughput makes SSDs totally unsuitable to server (read enterprise storage) applications. How does 40 I/O operations per second strike you? Yes, that’s 40, and it’s due to the inherent nature of NAND flash memory and its organization in SSDs. RAID helps some and hurts some – it spreads the load, improving performance, but makes sequential writes even less likely.
But these drive mechanisms aren’t meant for this kind of load – they’re meant for laptops and gamer PCs which heavily lean toward random reads, an area in which solid state disks excel. “Boot Windows in half the time!” (unless you’re comparing the SSD to an actual high-performance disk)
There’s also a little issue of reliability. Well, actually, let’s say longevity. You see, a NAND flash cell is still good for only about 10,000 writes before it becomes a lump. This isn’t much of a problem in a thumb drive or iPod since these don’t actually get written to all that much. Plus, all current SSDs include “wear leveling”, a strategy that maximizes write longevity by moving blocks (on write) to less-used flash cells. So a bigger flash device actually lasts much longer than a smaller one under the same I/O profile because all those extra cells can give their little lives to save your data.
This longevity issue isn’t just academic. Ask anyone with an Unslung NSLU2 booting from a flash drive (yeah, including me) and you’ll hear about failed thumb drives after a year or two of use. No big deal when you’re talking about a $15 item, but what about enterprise storage? I guess EMC and IBM wouldn’t mind forcing you to replace your enterprise storage media every year or two, but how will you feel about it? And what if the NAND was soldered to your motherboard when it failed? Makes Apple’s sealed iPod/iPhone batteries seem trivial, doesn’t it?
Then there’s the issue of cost. A 32 GB SATA SSD drive runs $400 retail, which is about how much you’d pay for a 300 GB SATA 2.5″ laptop drive (if you didn’t get a good deal). In other words, it costs about 10 times more than a comparable spinning drive on a per-GB basis. Lots of companies are investing in flash (iPod effect, anyone?) but NAND prices are not promising to overtake disks any time soon. So we’re left with an exceptionally flawed product.
Of course these drives have no cache – it’s irrelevant, say the manufacturers, in a solid-state device. But write cache might actually improve random write performance substantially, especially if it was backed by a super-smart algorithm to maximize sequential I/O in the same way that Network Appliance’s WAFL optimizes RAID-4.
And if this software also included the wear-leveling smarts, things would be even better. Imagine optimizing writes to kill a single NAND module quickly, sparing the rest of the array. Think tiered storage for longevity instead of cost – frequent write I/O operations go to the sacrificial cells and longer-lasting ones are destaged to the “permanent” ones. Makes media replacement much more palatable, doesn’t it? Add in some smarts and a sizable write-back cache to keep the really transient writes off the flash entirely and you might have something there.
As for cost, consider Mark Lewis’ recent posting about OLTP versus “web” data. He’s telegraphing EMC’s playbook for SSD – smart tiered storage that places small amounts of OLTP data on smart NAND and everything else on regular disks. Sounds workable to me!
One more point to make. Rumor has it, disk giant Seagate is thinking of snapping up memory specialist, Micron. If this isn’t a sign that solid state tech is becoming important to the storage component industry, I don’t know what it is!
So I’m still not too positive on SSD technology. It’ll always be more costly, and the current offerings are woefully inadequate. But I can see a way to make it work in enterprise storage, and I see signs that the big companies are trying to do just that. Wake me when the train arrives, ok?