An Attic Full of Boxes

So, what's going on with the SSD invasion? Is it the "Dave Clark Five" and not "The Beatles" as we all expected? Shouldn't the vendors be raking in the moolah as everyone segues from HDD to SSD? Doesn't look that way.

From a Wall Street point of view, it's all tanking. Who knew? Fusion-io, Violin Memory, and Nimble Storage all trade near lows. Oops?

Nimble is the most interesting case. It sells a Seagate-like hybrid drive, but it's market performance is shaky. Fusion-io parts are also server-centric. Which brings up the question: what of memory channel storage? Diablo Tech has been offering flash in DRAM slots for a bit. Again, a tad odd, and server-centric.

None of this would matter much, in that all of these vendors are after the same goal; speeding up data movement within a server. So, why did EMC scarf up DSSD? Read up this write-up, which is largely sound-bytes with EMC's Jeremy Burton. Sounds very much like Linus' observation is beginning to come to fruition. May be.

The DSSD machine is attached through the PCIe connector on a server. However, unlike PCIe devices from Fusion-io, the DSSD machine is connected remotely, rather than sitting inside each server. And so it acts as a pool of storage that fulfills some of the function of a traditional array, but without the latency introduced by traditional connections through fibre channel, ethernet, and interfaces such as "iSCSCI," says Burton.

What neither Burton nor the author address is the elephant in the room: Big Data and Big Analytics remain in the RBAR world, and for that purpose, 15K discs spitting out flatfile record after flatfile record is the bee's knees. Not to mention that analytics software (I'm talkin' to you, R) tends to do its iteration in its high-level code. A few milliseconds of latency won't matter. Refactoring those flatfile images to something more efficient, and with an engine which is inherently parallel (not, alas, Postgres), gains one a lot of speed. Except for corner cases such as median, the only reason quant software has been stuck with RBAR paradigms is that most practitioners still have Σ tattooed in their brains.

Interestingly, Violin has some quotes on its site's pages. I like this one.

If we had invented flash 50 years ago would we now package it up to look like disk drives? The short answer is "NO".
~Robin Harris, StorageMojo

While Harris doesn't, that I can find, display Linus' assertion, reading some of his posts might lead one to conclude that he agrees.

Organic Normal Form™ databases should be next on the agenda. Get all that silly, flabby, flatulent, flatfile Big Data under control.