http://bugzilla.suse.com/show_bug.cgi?id=1030310 http://bugzilla.suse.com/show_bug.cgi?id=1030310#c3 --- Comment #3 from Jan Kara <jack@suse.com> --- Even after disabling writeback throttling we can see noticeable effect of the commit mentioned in comment 0 on the numbers. Without the commit both read and write numbers are rather stable at: read[992]: avg: 18.3 msec; max: 881.4 msec read[998]: avg: 18.3 msec; max: 881.2 msec read[1001]: avg: 18.3 msec; max: 881.4 msec read[993]: avg: 18.3 msec; max: 881.0 msec wal[991]: avg: 0.0 msec; max: 0.2 msec wal[991]: avg: 0.0 msec; max: 0.2 msec commit[991]: avg: 53.9 msec; max: 928.4 msec With the commit read numbers are better but with large spikes like: read[16644]: avg: 7.8 msec; max: 369.4 msec read[16653]: avg: 7.8 msec; max: 369.6 msec read[16647]: avg: 6.6 msec; max: 369.4 msec read[16651]: avg: 7.8 msec; max: 369.5 msec read[16645]: avg: 7.8 msec; max: 369.5 msec read[16655]: avg: 7.8 msec; max: 20493.0 msec Write numbers are worse and with large spikes: wal[16642]: avg: 101.0 msec; max: 100993.5 msec wal[16642]: avg: 76.8 msec; max: 70830.4 msec wal[16642]: avg: 19.5 msec; max: 16180.1 msec commit[16642]: avg: 81.9 msec; max: 15718.7 msec wal[16642]: avg: 18.7 msec; max: 13445.0 msec ... commit[16642]: avg: 106.7 msec; max: 21362.7 msec wal[16642]: avg: 60.2 msec; max: 21059.4 msec wal[16642]: avg: 30.7 msec; max: 23651.2 msec -- You are receiving this mail because: You are on the CC list for the bug.