There is a battle shaping up in the enterprise storage industry. On one side are those who see flash storage as an upgrade for the disk drive, and on the other are those who see it as a cache between the CPU and the drive. It’s a fundamental difference of opinion - flash chips can be used either way, and each approach has its unique benefits and drawbacks.

The Register did a nice job of summing up the (late 2008) flash positioning of the various storage companies, and I recently posted a strategic look at this core issue. Note that some, like HP and Sun (and probably IBM), seem to have an end-to-end strategy, while others are firmly in one camp or the other. In the “not yet” column, apparently, are 3PAR, BlueArc (though they offer TMS RAM), Dell/EqualLogic, HP/LeftHand.

Update: 3PAR has joined the “disk” camp. Continue Reading »

With yesterday’s release of both an iSCSI version of it’s entry-level DS3300 and a complete, Microsoft Simple SAN-certified DS3400 solution, IBM is bringing the love to the … umm … well … compact (?) end of the storage market. See, we can’t call it “low-end” because these devices are decidedly not “low-end” in their functionality. And we can’t call it “entry-level” or “small business” because lots of established players buy this stuff. Modular and monolithic may describe some hardware, but it hardly differentiates the market. I refuse to start with tall like Starbucks (though you can get a super-tasty short cappuccino there!) So let’s just ape the car market and call it “compact”. There’s no shame in owning a compact car, especially with gas prices where they are, so why not get some compact storage to go with your green data center?

So what’s IBM doing this time? Well, they’ve taken LSI’s proven Engenio 1333 array technology (which they’ve used for a while) and turned on iSCSI functionality. IBM veers off course from the startups by not bundling snapshot and replication technology with their new array, however. This reduces the cost of entry but diminishes the impact of this new technology, since adding that software can easily double the price of this Ford Escort storage system.

IBM has also certified a complete Fibre Channel SAN solution with Microsoft, if you’re into that sort of thing. Their DS3400 can chat with an Emulex HBA and Brocade switch with quick setup and guaranteed compatability or your money back! (I made up that last part…) I’ve been pleased by Microsoft’s Simple SAN push in the past and think this is an excellent alternative to iSCSI for sites that aren’t ready to take the storage-over-Ethernet plunge yet.

Why care? Well, simply because this “compact” market is where the big action is in storage right now. Thanks to the VMware explosion, just about every smaller-than-10-TB shop is currently buying and deploying SANs right now, a fact that has warmed the hearts (and fed the sales people) at companies like EqualLogic, LeftHand, HP, Dell, and the rest for a year or so now. IBM was too early to market with an iSCSI array back in 2001, but has had nothing to sell since they axed the Adaptec-powered DS300/400 back in January. Welcome back!

Sure, hybrid drives are going nowhere fast in enterprise storage. But what about solid state disk technology? It’s been “almost there” for decades - anyone remember EMC’s solid-state Orion Atom (or was it Adam) array?

Now a handful of storage players are talking about SSD’s again, including some respectable names like LSI and Xiotech and some lesser-known outfits like Solid Data Systems and Texas Memory Systems. IDC is predicting mainstream uptake of the technology, too, but note that they’re mostly talking about the PC market, not enterprise storage.

I say that if SSD ever gains footing in the enterprise, it’ll most likely be in a virtualized hybrid system, acting like a mega-cache. Or maybe a non-RAM permacache, if you will! Of course, we’ve seen just how much uptake that feature has had, eh EMC? But hey, it could happen. Even though Apple is soaking up more than a quarter of the world’s NAND flash this year, maybe they’ll force prices down. Or maybe Samsung’s or Hynix’s PRAM will conquer the world!

Let’s get serious, folks. What good is SSD? It improves random access read performance over disk at the expense of longevity, throughput, and, well, expense! ReadyBoost looked like a nice application for flash memory, since the potential content was bounded and could fit on a flash drive, but it seems to have done precisely nothing for performance.

If enterprise applications could benefit from better random access performance, we’d be putting large amounts of memory in front of the disks already. Oh, wait, that’s right, we already do that! Modern enterprise arrays have gobs of cache, more than any SSD, and use it quite effectively. Remember the old RAID-5 penalty?

SSD just serves to remind me of one of those performance-tuning axioms I learned long ago: It’s always better to let an intelligently-designed system manage itself than try to second-guess it. This applies to all areas of system performance tuning, from filesystem and LUN layouts to cache tuning. And this is why permacache and things like it never caught on.

And it’s why SSD will continue to play just a bit part in the enterprise until it’s just as cheap as disk. Like that’ll happen anytime soon.

Edit: Looks like I got the name wrong - the EMC SSD was the Orion.  Introduced in 1989, this evolved into the Symmetrix in 1992.  I used one of these back at Texaco in the 1990’s.