1.012: flash drives in enterprise storage
Yesterday I presented a webcast on Enterprise Flash Drives to EMC's investment community, as part of Investor Relations' ongoing Tech Talk series. If you'd like to see it, the slides and replay are available on EMC.com here: Tech Talk for Investors: Flash Drives in Enterprise Storage, but only until July 25, 2008.
Chris Mellor has also reported about this presentation over on Blocks and Files in his "EMC on enterprise flash drives" analysis piece. (Thanks for noticing, Chris - glad you were able to participate).
On a related topic, Network World's Jon Brodkin takes a look into the growing market demand for Flash technology in his article titled Flash storage gets enterprise attention as prices decline. The article provides a fairly comprehensive look at what's going on in the realm of enterprise flash across both storage and server vendors, and it is definitely worth the read.
But I have to admit I was kinda surprised by this quote that Jon included from Michael Workman, president and CEO of Pillar Data Systems:
"The best use of solid-state disk is direct-attached, not in a shared network array. The reason for that is the latencies for solid-state disk are so low that putting it on a network to get at it actually makes the latency of the solid-state disk much worse than it could be."
IMHO, that's a pretty uninformed position to take, especially by the chief executive of an external storage vendor.
While there is indeed additional latency introduced by the storage network, it is not nearly as onerous as Mr. Workman seemingly would have you believe. If it were, we wouldn't be putting gigabytes of SDRAM inside of storage arrays - for as fast as flash is, the latencies of SDRAM is still orders of magnitude faster. And in fact, a Symmetrix DMX-4 can deliver a cached read hit from RAM anywhere from 4-100 times faster than the fastest available flash drive (depending upon the size of the I/O request). And in my presentation at EMC World demonstrated common database workloads where Symmetrix cache coupled with flash drives outperformed even server-based caches of the same size.
Following Michael's flawed logic, RAM caches would only be installed in servers and never in external disk arrays. The reality is that not only do the economics prevent that, as Bob Wambach (head of Symmetrix marketing) aptly explains earlier in the Network World article, but there is a very real performance benefit derived by buffering writes and pre-fetching reads in an Intelligent Cached Disk Array (a club of which admittedly Pillar ain't a member).
As most CIOs realize, When you have expensive resources, like RAM, flash or even disk drives, it is almost always more cost-effective to centralize and consolidate those resources so that their benefits can be shared across multiple systems. It is the reason why we have external storage in the first place, and that's why I'm surprised that Mr. Workman apparently doesn't know that the infinitesimal latency of a (well designed) SAN is far outweighed by the cost, performance and utilization benefits of storage consolidation.
Perhaps he learned that perspective from the Oracle of Sir Larry.