Quite Possibly The Biggest NAND Flash Experiment Ever
At about four o’clock today, I got one of the coolest surprises a guy obsessed with NAND Flash could ask for. A team of four guys - two from Facebook, two from Carnegie Mellon University - released what might be the biggest experiment ever conducted on NAND Flash reliability. In a four year experiment, the team gathered and analyzed all of the failure data from SSDs running on production Facebook data servers.
Let’s all take a moment to digest that fact - their experimental group of devices was all of Facebook’s solid state drive hardware, with the added bonus of all of Facebook’s user data as a test pattern. Can you get much bigger or better in terms of your sample size, or your test pattern?!?
The team made some pretty radical findings, not least of which are that specific configurations of solid state drives are vastly more prone to failure.
I won’t deny you the chance to draw you own conclusions on this - I still have yet to give it a third read and draw them for myself - but I had to take the chance to say how totally fucking awesome this is.
Take a look at the paper for yourself - you can find it here.