W dniu 19.01.2012 08:15, Stan Hoeppner pisze:
To demonstrate that fsync alone shouldn't be a factor here,
But it is. I've straced sendmail to "fsync" waiting lot of time. It was 80% or more of queue time.
Queuing 15K messages took 6 minutes 30 seconds on a single 7.2K drive, again while competing with the delivery agent for spindle time.
My previous 1 drive 1 cpu box was also quick.
If your kernel is using CFQ you may want to try deadline.
No change. Also try disable NCQ, set noatime, switch to reiser also - no effect either. Only change I found on 8 disk boot raid1 (no lvm) which appear even more slower.
But that alone isn't going to fix a 10x performance deficit. You've probably got multiple factors degrading performance.
Yes, you have right. But I found recently, that disk mounted on my server are slow 5.9K. My tests on in shows that they do fsync 1.5x-2x slower than 7.2K with quite often 5x-10x slower peak. Together with raid10, lvm, ext4, and much heavier load during delivery it may give effect I'm observing.
-- Konrad Rzepecki
