Hi Graeme,
Graeme Stewart wrote, On 25/11/08 15:19:
> I just looked and the jobs are very poor in CPU efficiency (15-25%).
> Yes, the jobs were reading directly using rfio.
>
That's not good. Looks like you will need a _much_ bigger switch and a
lot more network cards to support more analysis jobs. However, I'm
confused by your 15-25% numbers as this doesn't tie in with the
GangaRobot monitoring which says that Glasgow is running with ~50-60%
efficiency. I presume your 15-25% comes from the latet batch of jobs for
which the monitoring isn't available yet?
Ewan, the CPU/Walltime scale is just the % CPU efficiency.
> Event/sec is one of the outputs you'll see in the final analysis.
>
> Although the DPM servers were crusing - low load, excellent data
> output rates, the headnode was suffering very high CPU load. This is
> surprising as the headnode should only be contacted for the open step
> and it hands off to the disk server.
Yes, the headnode should only be seeing the initial connection as the
client tries to access the file in the DPM namespace, so I'm confused as
well.
At Edinburgh we've been running a lot of user analysis jobs against our
DPM for many months now and I can't say that I've really seen a huge
load on the head node. I'll need to check.
Greig
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
|