The Elephant, The Blind Men And Fusion IO

Last week, Enterprise Strategy Group's Steve Duplessie opened a real can of worms with his blog post "Why I'm Bearish on Fusion IO." By questioning Fusion IO's prospects, Steve insulted the popular girl at the prom, and several other industry observers played the role of the football team coming to Fusion IO's defense, with blog posts like "What Storage Folks Don't Get About FusionIO" and "Steve Duplessie Hates Fusion IO".  Of course, all of this chatter is in response to Fusion IO's filing for an initial public offering in March.

Fusion IO has done some brilliant things--including hiring geek god Steve Wozniack to be its chief scientist and making it known that Facebook has bought several million dollars' worth of its kit--earning the company more mindshare than its roughly $100 million dollars in annual sales would otherwise deserve. It's reached the point where many of us use Fusion IO as shorthand for PCIe flash, even when talking about applications like NetApp's Flashcache (formerly known as PAM).

Steve makes some good points in his post, starting with the fact that, as a maker of SSD PCIe cards, Fusion IO has lots of competition and lives at the mercy of the actual flash makers. If I were in the market for such a product I could call LSI, OCZ, TMS, Virident and 20 others. Should Micron, Intel, Samsung or other flash makers want to play in this space, they'll have a cost advantage over Fusion IO due to vertical integration.

Then there's the issue of PCIe flash being inherently DAS and therefore counter to the trend to shared storage. Given that the trend to server virtualization depends on shared storage for features like vMotion, the shared storage trend is likely to continue.

The naysayers--at least those who don't accuse Steve of pooping on Fusion IO because they're not an ESG client--say Steve thinks too much like a storage guy. Spinning rust is so 20th century, and the whole concept of treating flash like disk is a bad idea. After all, the coolest companies are using Hadoop-like distributed DAS applications and don't need those nasty old SANs.The whole affair reminds me of the parable of the elephant and the blind men. One felt the side and thought it was a wall; the second felt a leg and thought it was a tree; the third interpreted the trunk as a snake.

With his enterprise focus, Steve looks at Fusion IO and sees a niche product that addresses I/O latency if, and only if, that problem can be addressed within a single server. The Web 2.0 guys see a broader solution. The problem is that the NoSQL Hadoop world lives with "eventual consistency"--it doesn't really matter if different Facebook users on different nodes see drunken frat boy photos a few minutes apart. The corporate world, however, requires consistent data. If I place an order at Amazon and the website says the item is in stock, it had better actually ship today.

Once again, the old horse player's motto applies "Different horses for different courses."

Additional Links: