On Wed, 2014-07-23 at 16:20 +0000, Sue Ciani wrote: > Has anyone implemented solid state drives on their database servers? If so, > what was your experience? Did it increase response time? Did you put them > only on your database server?
We implemented on SSD from the start in March 2011 for both the frontend and backend. We have both as VMs rotating across three physical hosts (one is always cold unless we are testing a new Evergreen version by running both front and back on the spare) which spreads the wear a bit. But while I was a bit worried about the stories we all hear about SSD wear, not a problem. The lowest number I'm seeing right now for Media Wearout Indicator is 97. The flash will almost certainly outlast the servers they are mounted in. Now for the downside. Don't know if it is something in the VMs (although I have even tried migrating a front to bare metal as a test and it didn't help) Perhaps it is something inherent in the multiple layers of abstraction PINES needed to scale beyond what a single server could do back at the turn of the century, general lack of performance inherent in XUL + Javascript on the Staff Client or what but I have never been impressed by EG performance. Even on an unloaded network (after close) requests involving large numbers of items take far longer than they 'should.' Direct queries with PGAdmin or Perl scripts are plenty fast since the whole DB fits in ram. So don't go buying a stack of SSDs hoping to see a big performance gain, rotation speed apparently isn't the bottleneck.
signature.asc
Description: This is a digitally signed message part