My copy of the archive must have been corrupted. I repeated my data
collection on ds1, brought the data to Apollo and was able to use pmval. I
did try your command "pmval -z -a test2.ds1 kernel.all.cpu.idle" without
success on the old data collection. If I get a chance I'll try the sum and
md5sum commands. Thanks for your help.
Of course now that I can collect and extract the performance data the
question becomes what does it mean. Most of the metrics have fairly
self-explanatory descriptions but others are not so clear to me. Can you
point me to a url or some discussion group that has definitions of the
metrics generated by PCP?
Finally, we have an installation of SGI machines for which I will need to
measure the CPU, Memory, Disk and I/O usage. I know PCP is available
commercially for IRIX machines and would like to get a price quote for PCP
for IRIX. Could you forward this e-mail to the appropriate SGI sales
person and ask them to call me at 805-348-2124 or respond to this e-mail?
Thanks
George Rapacki
Lockheed-Martin
> ----------
> From: Mark Goodwin[SMTP:markgw@xxxxxxx]
> Sent: Sunday, June 02, 2002 5:09 PM
> To: Rapacki, George R
> Cc: Linux Perf CoPilot Majordomo
> Subject: Re: How to do data reduction on a machine different from the
> data col lection machine?
>
> On Fri, 31 May 2002, Rapacki, George R wrote:
>
> > I have used pmlogger to collect performance data on our ds1 server. I
> can
> > use pmval on the ds1 server to see the metric values. When I copy the
> > archive files and the pmns to another machine with PCP (Apollo) and try
> to
> > run pmval or pminfo or pmdumplog I get an error, specifically:
> >
> > pmval -z -a test2.ds1 -n ./pmns/root kernel.all.cpu.idle
> > pmval: Cannot open archive "test2.ds1": Illegal label record at start of
> a
> > PCP archive log file
> >
> > Any suggestions on what I'm missing?
>
> Firstly, you usually don't need to explicitly specify the pmns file
> when replaying an archive because the pmns is included in the archive
> itself. Does the folowing work: pmval -z -a test2.ds1 kernel.all.cpu.idle
>
> Otherwise, this error occurs when an archive has been corrupted,
> specifically when the archive label is corrupted. Check the three copied
> archive files on Apollo are byte-for-byte identical to the originals
> on ds1 (use sum or md5sum). If that's OK, then please post the
> archives to me or if you can't do that, send the pmdumplog -L output.
>
> thanks
> -- Mark
>
|