On Fri, 8 Feb 2008, Paul Goyette wrote:

> I've actually been seeing this for a while - at least since 4.99.39 ...
>
> Occassionally it seems that disk I/O activity just stalls.  currently I've 
> got a dump(8) running on the filesystem that contains /usr/src and /usr/obj . 
> The dump has been running for more than 5 hours now, and it making extremely 
> slow progress.  Sometimes the dump takes 9 or 10 hours before it finishes! 
> The hard drive activity light blinks about once a second - it should be on 
> solid, since the machine is doing nothing else.
>
> The dump is running under control of /usr/pkg/sysutils/amanda but I don't 
> think amanda has anything to do with the problem.  The dump command is being 
> piped to gzip(1) and the gzip output is directed to a network socket 
> (amanda's 'dumper' is on the other end of the socket and writing things to a 
> disk file on a different physical drive).
>
> I've seen similar "stalls" when gzip is trying to uncompress a file located 
> on an NFS-mounted file system.
>
> And recently (since installing 4.99.53 yesterday) I'm occassionally seeing 
> long waits for pine to access my mailbox - up to 1 minute or more in one 
> case.
>
> Has anyone else seen similar behavior?  Any clues on how to track down the 
> cause of this slowness?

Just an added datapoint here, which may (or may not) exonerate NFS...

I just tried to HUP a process (and mistyped the PID), and the

 	kill -HUP x

command took more than 10 minutes to give me a new shell prompt!

----------------------------------------------------------------------
|   Paul Goyette   | PGP DSS Key fingerprint: |  E-mail addresses:   |
| Customer Service | FA29 0E3B 35AF E8AE 6651 |  paul@whooppee.com   |
| Network Engineer | 0786 F758 55DE 53BA 7731 | pgoyette@juniper.net |
----------------------------------------------------------------------