A bit of bottle neck.
Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s rkB/s wkB/s avgrq-sz avgqu-sz await svctm %util sda 0.38 176.63 70.32 78.26 813.46 2044.82 406.73 1022.41 19.24 0.40 19.17 4.04 60.07 sda1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 5.28 0.00 23.61 19.33 0.00 sda2 0.38 176.63 70.32 78.26 813.45 2044.82 406.73 1022.41 19.24 0.40 19.17 4.04 60.07 dm-0 0.00 0.00 70.71 255.60 813.45 2044.82 406.73 1022.41 8.76 2.90 8.87 1.84 60.10 dm-1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 8.00 0.00 64.20 11.38 0.00
Try setting the scheduler to 'deadline' and see if the queue sizes shrink.
No raid1? Besides adding redundancy, it can help with read performance. I would probably put the mail on a raid 10 though if I had 4 disks to do so.
-Ross
Like this:
kernel /vmlinuz-2.6.9-78.0.8.ELsmp ro root=/dev/VolGroup00/LogVol00 elevator=deadline
And this change will be for System Wide.
Doing this among other things helped quite a lot. My iostat -x %util dropped from ~ 60% to ~ 22% now. Of course at same time I updated from dual-core 2Ghz CPU to quad-core 2.4Ghz. Also swapped motherboard from one with an "Intel ICH9R" Southbridge to a "Intel ICH7R" since I heard CentOS 4.x did not have drivers in kernel for ICH9R.
Matt