Yep the software poller I was using, make a lot of bulk walk on the
entire part of the Mib (I was using observium and it had a so poorly
coded poller).
Other software made more intelligent polling using cache and atomic get
(like cacti).
Anyway even a bad snmp poller should not kill the control plane of a
router/switch.
Junos should have implemented software qos, leaving some cpu time for
critical daemon (rpd, lacpd, etc..) or even better use a real time os
(like some other vendor do)...
The fun thing is that manually nicing the snmpd process help. I wonder
why juniper does do that by default ?
--
Raphael Mazelier
Le 28/04/2016 à 22:02, Aaron a écrit :
About snmp...
Is your Management station walking the entire mib tree of your devices ?!
I've experienced over the years that when I management station is NOT tuned
appropriately, it WILL spike the processors of your routers/switches during
the polling cycle, like every 15 minutes or so, whatever it is set to.
I'm not sure this is your problem, but it could be.
Aaron
p.s. it's probably dumb/unnecessary to have a snmp mgmt. station hit all the
OIDs on the network devices.... since you probably don't want/need that much
info queried anyway.
-----Original Message-----
From: juniper-nsp [mailto:[email protected]] On Behalf Of
raf
Sent: Thursday, April 28, 2016 11:03 AM
To: [email protected]
Subject: Re: [j-nsp] LDP flaps specifically present on ACX Juniper routers
(ACX4000 and ACX1100)
Hi,
Is your RE cpu utilisation is OK or not ?
I have similars issues on EX during snmp polling, which are eating all the
cpu time.
_______________________________________________
juniper-nsp mailing list [email protected]
https://puck.nether.net/mailman/listinfo/juniper-nsp
