Re: R: Cricket Performance

> -----Original Message-----
> From: [email protected]
[mailto:cricket-users-
> [email protected]] On Behalf Of Angelo Deflorio
> Sent: Thursday, June 08, 2006 2:57 PM
> To: 'Levin, Craig A.'; [email protected]
> Subject: [cricket-users] R: Cricket Performance
> 
> I've been monitoring about 600 routers some years ago, having non so
much
> CPU power on my cricket servers.
> Having so much target to collect you have to use multple collectors,
with
> high CPU usage and cricket web interface slow down.
> 
> My solution was to split the two basic cricket functions (collector
and
> grapher) and to spread the collector function across a pool of
servers.
> In this case I used:
> - one front end Linux server (RH70), apache equipped, hosting the
cricket
> configuration and cricket data exported via nfs, used to browse
collected
> data.
> - three back end Linux servers (RH70), linking the same front end nfs
for
> config.db and rrd database, each one collecting its portion of data
(about
> 200 devices) all the time.

I have the same kind of configuration but without NFS. I have 5 data
collectors each collecting stats on about 1000 devices each that I
periodically rsync config and data information between a central host. I
am definitely planning on moving to the NFS model and was concerned that
there might be file-locking issues, etc. Glad to hear that it worked out
for you. Do you happen to track bandwidth utilization between your
collectors and the central NFS host?

Thanks,

--
Marc