My original message was:
"Not long ago, I installed DU 4.0D patchkit 3 on our two Alpha Server 4100s,
> and starting the next morning we were experiencing what appeared to be
> hardware problems. We were getting CMD timeouts and Misc adapter errors
> logged at about the same time for three days straight. Only once did this
> cause any AdvFS panics and cause filesystem to go unavailable, but it
> really
> had us tied up in knots.
>
> To make a long story short, we discovered that the problem was being
> caused
> by the defragment utility that UNIX so generously inserted into root's
> crontab for us. I knew that happened whenever you do an O/S install, but
> not
> a patch install. There were 14 defrag processes running at the same time
> on
> our database server and 5 on our other server, all of which were loading
> our
> shared bus down so much that no other I/O could get through. These
> processes
> took around four hours to finish, and if it weren't for the fact that we
> were in the office trying to track down the problem at 1AM, we wouldn't
> have
> found what the problem was.
>
> My question is, when the defragment utility is running, is it okay to kill
> it or will that cause serious problems?"
>
>
I got several responses on this one, from most of whom are afraid to stop a
defrag once it starts. alan_at_nabeth.cxo.dec.com wrote:
Defragment is supposed to gracefully deal most signals. Since
the majority of the work happens deep in the kernel, I suspect
that even a SIGKILL is unlikely to perturb it when it is in the
middle of something important.
Another option is to specify just how long each defrag process is allowed to
run. In this example, the process will self terminate after 60 minutes of
execution:
/usr/sbin/defragcron -p -T 85
Thanks to all who responded.
-Stephen Spalding
Received on Thu Jun 24 1999 - 21:48:50 NZST