Stephane Bentebba stephane.bentebba@fps.fr writes:
for me, the disk activity is directly related to the scrub process scrub consist of read and re-write of what was baddly read so.
I'm not clear whether Blake's sysstat extract is from during the scrub or after it had finished. The disk read rates are tiny for an F760 doing a scrub: even the "around 9600bB/s" (should that be kB/s?) sounds very low. A scrub won't do any significant writing unless your discs are in a dire state! (or its an upgraded volume and this is the scrub that finishes off doing the zone checksums).
The (very) significant performance issue that the sysstat shows is the "cache age" being 0s. A scrub doesn't normally have much effect on the cache age.
So I'm left to speculate about
This system has been online for 400 days, 11:40 2000225709 NFS opsand I think it might just be time for a reboot.
There could, I suppose, be something awful that happens after either 400 days or 2 * 10^9 NFS operations! (How does Blake manage so exactly 5000000 NFS operations a day, I wonder?) I admit to never achieving that much between reboots on any filer here.
(Munge through archived logs... I see I made 292 days and 1296 million NFS ops back in 1998 but that was on an FAServer 450 and FASware 3.x. In more recent years I don't seem to have got above 133 days and 731 million NFS ops. Must be the frantic desire to keep up to date on ONTAP versions, I suppose.)
Chris Thompson Email: cet1@cam.ac.uk