On Wed, 2003-05-14 at 09:18, Douglas Eadline wrote:
> Hey Nick,
>
> Quick question, I just upgraded ganglia to version 2.5.3.
> on a 300 node cluster I have seen gmetad slowly load the
> monitor node so that it is almost useless. Have you seen
> any thing like this? I am going to submit to
> the list, but I was wondering if you had any experience with this.
Hey Doug--
We do not have a single source that is 300 cpus, but between 8
clusters, we are running about 600 cpus total. Our gmetad server gets
the info from a gmond running on each head node. The gmetad server is a
quad P3 550, and we have placed /var/lib/ganglia on it's own scsi disk.
The separation of disks seems to have really helped the interactive feel
-- I have a feeling that the interactive use goes to arse because it is
waiting to read data from the disk. We have also mounted the disk with
noatime to help speed up the accesses as well. We do see almost a
constant load on that disk due to the rrd's being updated.
Nic
--
Nicholas Henke
Penguin Herder & Linux Cluster System Programmer
Liniac Project - Univ. of Pennsylvania
