On Fri, Nov 22, 2013 at 02:07:48PM -0800, Ned Bass wrote: > Thanks, it's good to get confirmation that this was the problem. I will > post updated results with the fix in place. > > Ned
New benchmark results with patch 4347 are now available: https://github.com/nedbass/zfs/wiki/ZFS-Write-Throttle-Patch-Benchmarks The patch completely erased the problem I was seeing at high thread counts. The new IO scheduler is a clear winner for many of the tests, and performs equally in the others. This is very encouraging! I also corrected the anomalous jumps in some of the performance curves of the original read tests, which were due to a flaw in my test scripts. For comparison, the original results are still available here: https://github.com/nedbass/zfs/wiki/**BROKEN**----ZFS-Write-Throttle-Patch-Benchmarks Ned _______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
