[Beowulf] SSDs for HPC?

[Lockwood, Glenn]

On Apr 7, 2014, at 1:40 PM, Ellis H. Wilson III <ellis at cse.psu.edu> wrote:

> I guess I disagree with the previous poster that saving by going the commodity route, which, by the way, is not pennies but often upwards of 50%, is always bad.  It really depends on your situation/use-case.  I wouldn't store permanent data on outright commodity SSDs, but as a LOCAL scratch-pad, they can be brilliant (and replacing them later may be far more advisable than spending a ton up front and praying they don't).

The problem we faced was unique to production-scale resources.  Buying commodity desktop SSDs is fine if you are just buying a handful, but we were having major reliability and performance problems right out of the box at-scale.  The cost of time lost to swapping and RMAing tons of junk SSDs, which you will find in the cheap end of the SSD market, starts to add up.  Although the technology landscape may have improved since our last big deployment, anyone who goes this route should plan on hiring students to do the monkeying around.

In our case, we weeded out most of the cheapo SSDs in testing before production hardware ever hit the floor, but our vendor made the decision to improve their cost margins by going for the minimum spec that would make acceptance.  It turned out that the advertised reliability/performance did not match the delivered product, and they had to replace hundreds/thousands of SSDs with Intel's enterprise line.  At the time, Intel's hardware was the only technology that could pass acceptance testing.

Glenn