On Thu, Sep 18, 2003 at 12:53:01PM +1000, Jeremy Field wrote:
> Thanks for your exceedingly rapid reply...
>
> >> I am using linux 2.4.21 with xfs 1.3 on debian-unstable. Since I
> >> upgraded from 2.4.18-xfs-1.1, my xfs partitions have been showing
> >> pulses of activity every five seconds, I think because of xfssyncd.
> >> The hard drive physically seeks every five seconds too.
> >>
> >> I have reviewed this mailing list and elsewhere for clues, but can't
> >> work out either:
> >> a) whether this behaviour is normal, or
> >> b) how to change the interval.
> >
> > A sysctl variable controls this: /proc/sys/fs/xfs/sync_interval
>
> Ok I hadn't realised that's where it lived. Mine's set to "3000". Does
> this value rule out the diagnosis I've made?
thats every 30 seconds, so its unlikely to be xfssyncd.
> > >>From Documentation/filesystems/xfs.txt...
> >
> > fs.xfs.sync_interval (Min: HZ Default: 30*HZ Max: 60*HZ)
> > The interval at which the xfssyncd thread for xfs filesystems
> > flushes metadata out to disk. This thread will flush log
> > activity out, and do some processing on unlinked inodes
> >
> > What makes you suspect xfssyncd? It could also be some arbitrary
> > process deciding to write something out every few seconds of course.
>
> The reason I suspect xfssyncd is because it happens simultaneously on
> all of one drive's xfs partitions together (ie /{,usr,var,home}), and
> not on its non-xfs partitions. Additionally, this behaviour is new since
> I upgraded to xfs 1.3.
hmmm.
> It doesn't actually seem to be happening on another drive which has a
> mounted xfs partition. Some other possibly relevant settings:
>
> $ cat /proc/sys/vm/bdflush
> 30 500 0 0 500 3000 60 20 0
> ^ interval in 1/100s between kupdate
> flushes -- could this be my problem?
possibly. worth fiddling with that knob and seeing what happens,
anyway.
cheers.
--
Nathan
|