[svlug] More links about ssds

Sarah Newman newmans at sonic.net
Thu Jun 2 10:40:03 PDT 2016

(combining multiple responses)

On 06/02/2016 12:56 AM, Rick Moen wrote:
> Several _actual_ experts on SSDs (includign Sarah Newman) were kind
> enough to help from the audience, and identified a couple of the many
> additional subtopics that might have been included.

Hah. Thanks, but my level of knowledge is only a few days more of research :) I haven't been using SSDs in production at large scale.

On 06/02/2016 09:48 AM, Rick Moen wrote:
> Sarah made the interesting point that she's seen SSDs fail
> catastrophically with very little advance indication of impending doom,
> which might take one by surprise if one is used to HDs that _most_ often
> signaling frantically that they're on their last legs and you should be
> doing that final backup instanter.

It was not from personal experience. I was talking about this article from the same source Marc Merlin referenced:
http://techreport.com/review/26523/the-ssd-endurance-experiment-casualties-on-the-way-to-a-petabyte/2 "When we checked on the SSDs after 900TB of
writes, the 840 Series was still functional, and Samsung's own SSD Magician software gave it a clean bill of health. The 840 Series didn't make it to
a petabyte, though. It died suddenly in the last leg, without any preceding SMART warnings."

RAID and SSDs are another interesting topic. I think this is a reasonable overview though it's from 2011:

I get the impression it's more likely for SSDs with the same age and workload to fail concurrently than with hard drives, though I don't know how
serious that risk is in practice if the SSDs you're using aren't designed to self-destruct after a certain number of wear cycles.

It doesn't hurt to mix manufacturers and/or hours of use within a single array. Unevenly distributed parity (if using parity-based RAID) is another
approach to reduce the possibility of having multiple drives fail concurrently: http://www.cs.yale.edu/homes/mahesh/papers/eurosys10-diffraid.pdf


More information about the svlug mailing list