On Sep 27, 2010, at 9:54 PM, Ross Walker <rswwalker at gmail.com> wrote: > On Sep 27, 2010, at 8:16 PM, Tom Bishop <bishoptf at gmail.com> wrote: > >> Here are the iostats: >> >> >> Device: rrqm/s wrqm/s r/s w/s rsec/s wsec/s avgrq-sz avgqu-sz await svctm %util >> sda 0.15 2.47 0.41 0.82 13.01 26.36 31.97 0.01 6.98 1.01 0.12 >> sda1 0.02 0.00 0.00 0.00 0.04 0.00 24.50 0.00 5.38 4.82 0.00 >> sda2 0.01 0.00 0.00 0.00 0.03 0.00 37.79 0.00 6.77 5.85 0.00 >> sda3 0.12 2.47 0.40 0.82 12.93 26.36 31.98 0.01 6.96 1.01 0.12 >> sdb 1.48 0.00 315.21 0.01 40533.39 0.75 128.59 26.94 85.45 2.80 88.24 >> sdb1 1.47 0.00 315.21 0.01 40533.30 0.75 128.59 26.94 85.45 2.80 88.24 > > Average queue size of 26.94 requests, average wait time of 85.45ms, service time of 2.8ms ain't bad, but means the sequential IO is randomizing and backing up the IO. > > Chances are this is probably a 4k sector drive and the partition's alignment crosses a 4k page causing double reads. Better to start partitions on sector 2048 instead of 63. > > Am I correct on these? > > If so I'd break the RAID re-partition and resilver it. I was wrong about the sector size, it's regular 512 byte sectors. It still makes sense to look at the partition offset, but I would also look at the cabling too. -Ross