[j-nsp] Graphing VCP Backplane (possible jnxUtil SNMP bug?)

[Charlie Allom]

Hi,

Has anyone seen high CPU with many snmpgets?

jcs:execute() in the SLAX helped.

I've come across another performance issue - running through the jnxUtil
MIB - it drives the load up significantly:

  http://d.pr/VR4R (spot the polling)

I've included my collectd config here:
http://capslock.playlouder.com/collectd.conf

It's all 64Counters.

You can see that sw0 (high cpu) has more ports than sw1. sw1 has little
to no CPU. Both are 10.4R5.5

In both cases, running the script every 5 minutes creates no noticeable
load - only in polling.

I can certainly drive the CPU load up if I poll every few seconds on
either VC stack :)

Cheers if anyone has some information,
  C.

On Wed, Oct 19, 2011 at 07:02:12AM -0700, Curtis Call <ccall at juniper.net> wrote:
> You'll want to use jcs:execute() instead of jcs:invoke(). Take a look at page 69-71 of the Mastering Junos Automation book for details on the performance difference when executing a lot of RPCs:
> 
> http://www.juniper.net/us/en/community/junos/training-certification/day-one/automation-series/mastering-junos-automation/
> 
> 
> > -----Original Message-----
> > From: juniper-nsp-bounces at puck.nether.net [mailto:juniper-nsp-
> > bounces at puck.nether.net] On Behalf Of Charlie Allom
> > Sent: Wednesday, October 19, 2011 7:11 AM
> > To: juniper-nsp
> > Cc: Christoph Blecker
> > Subject: Re: [j-nsp] Graphing VCP Backplane (slax help needed)
> > 
> > On Tue, Nov 09, 2010 at 04:30:28PM +0100, Phill Jolliffe
> > <phill.jolliffe at gmail.com> wrote:
> > > If you can find a counter for the vcp throughput then you can
> > populate
> > > the "utility mib" with the value and snmp poll and graph it.
> > 
> > I've done just this. It took me about 4 hours of learning SLAX.. enough
> > to patch this together:
> > 
> >     http://capslock.playlouder.com/check-vc.slax
> > 
> > It actually only works as a proof of concept.
> > 
> > On a 2-member EX stack with little traffic, the load is high but it
> > only takes a minute to run.
> > 
> > On a 3-member stack running decent traffic, the load grinds to 1.7 and
> > the CPU is very busy. On anything more than 3*24T I would imagine the
> > script is unusable!
> > 
> > The output is nothing like normal interfaces.. you'd need to craft your
> > own poller with collectd or something to watch the traffic.
> > 
> > Can anyone help me with SLAX to maybe bring the load down? I am sure I
> > am doing-it-wrong.
> > 
> > Regards,
> >   C.
> > --
> >  +442077294797
> >  http://mediasp.com/
> > _______________________________________________
> > juniper-nsp mailing list juniper-nsp at puck.nether.net
> > https://puck.nether.net/mailman/listinfo/juniper-nsp

-- 
 +442077294797
 http://mediasp.com/