>> 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