EMC's solid-state play begs for benchmarks

Putting SSDs in Symmetrix looks good on paper, but proof of performance-to-cost ratios are in the testing

I would bet dollars against pennies you didn't miss the EMC buzz about SSDs (solid-state drives) in Symmetrix. The vendor carefully orchestrated the announcement in hopes of capitalizing on the most interesting innovation to its portfolio in a long time.

[ MarioApicella's column is now a blog! Get the latest storage news from the Storage Adviser blog. ]

How carefully? As The Register notes (with evident disappointment), trade press reporters were not the first to get the news: "The publicity stunt nature of EMC's announcement was made all too clear by the vendor leaking word of the SSDs to the Wall Street Journal."

Flirting with Wall Street? You bet. And I'm not surprised EMC is trying to squeeze all it can from its first step into SSD, however limited that step may be.

More on that "limited" comment below, but first I would like to direct your attention to an article published years ago on dbazine.com comparing the query performance of an Oracle database on spinning drives vs. SSDs.

Obviously, the difference is huge, days vs. minutes, as it should be when you compare radically differing media. My point, however, is that vendors such as Texas Memory Systems, which provided the SSD gear for that test, have long been suggesting that, when targeting top performance, using SSDs can be paradoxically less expensive than using spinning drives.

I wouldn't be surprised if TMS and vendors such as Gear6 and XIOtech, both of which recently made SSD-related announcements, feel vindicated that a big gorilla like EMC has come around.

Hype aside, to find out how EMC's new SSDs can improve the performance and economics of its Symmetrix DMX4, check out this white paper.

First, an SSD can sustain 30 times more IOPS than a 15,000-rpm drive, allowing you to achieve the same performance using significantly fewer drives.

Another interesting point from the paper: EMC suggests that, using its flash drives, application response times will not only be shorter (say, 1ms), but also more likely to remain close to those values even under a heavy transactional load.

I don't doubt that; in many systems, bottlenecks will move from the usual suspects -- the spinning drives -- to other components, such as the array controllers, the application server, or the network pipes.

EMC is not talking price yet, but it suggests that a price-per-IOPS comparison would prove its flash drives to be on par with its spinning drives. Moreover, you can retrofit an existing DMX4 with the new SSDs, if you first bring the array software up to snuff.

Making the update will no doubt be pricey, perhaps prohibitively so. But according to EMC, an SSD configuration will deliver a much better watts-per-IOPS ratio, allowing you to achieve comparable performance using 98 percent less energy.

On paper, the drives look appealing. But how many does EMC expect to sell? My guess: not a lot, but the expected high price tag will make for a worthwhile incentive for EMC's salespeople.

Assuming, of course, customers can be persuaded that those performance-to-cost ratios are indeed accurate.

It's one thing to make the performance of an array with SSD drives a major selling point; it's another to publish benchmark results and submit the arrays to independent, peer-based reviews. I can always dream, can't I?

Join me on The Storage Network with questions or comments.

Join the discussion
Be the first to comment on this article. Our Commenting Policies