On Dec 8, 2008, at 6:51 PM, Matt <lm7812 at gmail.com> wrote: >>> 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. > > I have googled this and having a bit of trouble figuring how to change > it under CentOS 4. > > http://www.wlug.org.nz/LinuxIoScheduler > > Does not seem to work on CentOS 4. > >> 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. > > I plan on moving to faster disks and RAID 1 down the road. Just has > not happened yet. Setting scheduler is global in C4 it can be set as a kernel option with a scheduler=deadline in grub. -Ross