On Wed, 23 Apr 2008, Toshio Kuratomi wrote: > Mike McGrath wrote: > > latency and throughput measure different things though, we'll have to test > > multiple things. > > > Very true. From the data we've gathered at tummy, though, I'm not nearly as > concerned about throughput as latency. Latency makes every connection to the > database or to a proxy server that's not in the local colo much more expensive > no matter how large or small the data returned. With tummy, in fact, I know > that making fewer requests trumps the amount of data returned by a large > margin. > so here's a simple and non-dedicated test against our current environment. In the first tab I hit app2,3,4,5,6. App5 and 6 are not in PHX. One thing you'll notice is that our remote hosts have a much higher max request time though 90% of the requests were handled in the same ballpark as the local machines. Also, for some reason, app5 is responding more quickly then app4. initially I'm suspecting this is because app2,3,5 are i386 and app4,6 are x86_64. I'm going to run some more tests and verify that. Also in the longer tests, where I hit app servers directly with concurrent requests over time, aside from the drop outs (still looking in to wtf happened there) I'd say we're in pretty good shape. Way way beyond capacity for normal days. I'll look at what the last release looked like but mirrormanager also was no problem for the last release. -Mike
Attachment:
Benchmarks.ods
Description: application/vnd.oasis.opendocument.spreadsheet
_______________________________________________ Fedora-infrastructure-list mailing list Fedora-infrastructure-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-infrastructure-list
