Marc Bevand <m.bevand <at> gmail.com> writes: > > One of my machines with 4 SATA disks is resyncing a sdb, which is part of a > 2-way raid1 MD array with sda, but is doing so _very_ slowly at a speed of > 5-10 MByte/s. The machine is idle (no CPU, no fs activity on top of the > array) and dev.raid.speed_limit_max has not been modified (left to the > default value 200000). The output of iostat doesn't make sense to me: why > are there 2-3 seconds periods of sdb being 100% utilized with an avgqu-sz > of 4 but an avgrq-sz of 0 ?! What does this mean, that 0-byte I/O requests > are being issued to the disk? I think the avgrq-sz values of 0 are probably a minor bug in iostat when I/O requests take more than 1000ms to execute. Anyway I fixed my problem. sdb was really having a hard time. It had loose disk tray screws, which were causing excessive vibration, which in turn were causing disk head seek errors, which degraded I/O throughput... This explains everything. I should have paid more attention to the SMART data (high Seek_Error_Rate.) This experience reminds me of the blog post from a Sun engineer who demonstrated that shouting at a ZFS disk array produced sound waves powerful enough to cause vibrations and I/O latency spikes... -mrb -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
