Hello
> Installed that. Processes frequently using disk seem to be:
> jbd2/dm-0-8
> dhcpd -q
> kworker/0:0
> kworker/0:3
>
> jbd2 is of course the journaling process for my RAIDs. But who's I/O is
> it journalling?
>
> dhcpd -- whi does it have to read disk every second? connctions aren't
> being made that oftern, are they?
>
> And what are the kworkers all about? Somthing kde-ish?
So a "ps -eF" should show you that both kworker and jbd2 have zero
pages (SZ). That generally means that they are part of the
kernel.
My suggestion: Stop dhcpd for a while and see if the click goes away.
I'd do a "kill -STOP pid", then a sync, then listen for a few
seconds, then do a "kill -CONT pid".
My suspicion: It isn't dhcpd. I have two candidates:
* Hardware unhappy. Use smartctl to look at the disk
statistics. A nonzero reallocated sector count is or used to be
quite a red flag. A new install would have triggered loads
of writes which could have generated bad sectors.
* New filesystem type which decides it is necessary to sync all
atime updates to disk, or something similar. Try
tricks involving mount: Possibly "mount -o remount,noatime /some/path"
or even (after having killed processes which have files open
for writing) "mount -o ro /some/path"
I have lots of angry words for people who write userspace
code, drivers and filesystems which actually aren't
idle when they are idle. Good code blocks in a read(),
select() or epoll() when idle. Good drivers wait for an
interrupt, and good hardware makes this possible. Thesedays
I go around and strace processes "strace -p pid" and disable
those which don't.
There are other less likely things to consider: Hard
disks are computers in their own right (often dual
core ARMs, soon some might be RISC5s). They can
do things like thermal recalibration which can make
noise.
regards
marc
_______________________________________________
Dng mailing list
[email protected]
https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng
