Jiri Dvorak
2009-03-27 07:56:45 UTC
Hi,
are there any recommendations / experiences with high values for the
-mp parameter for the async collector? Somewhere in the code it says
10 to 20 are reasonable values. But now I've run tests with mp 10000
and see no issues so far.
We currently poll some 5000 targets on (mostly) Cisco routers and
switches. Some with 1min polls and some with 5min. This works fine,
ideally the collection is done in less than 30sec.
But if there is an outage in the network, many devices won't reply,
causing timeout delays in snmp collection. If you collect with -mp 10,
you risk to have a lot of these timeouts in sequence, and the
collector won't finish before the next polling cycle.
If you have mp higher than the actual number of targets, all async
requests are sent out right away, and there is only one waiting for
timeouts in the end.
Are there any long-term experiences with such high mp settings? I have
it running, but no big network outages yet. And the collection server
(Linux x86) doesn't seem to have any issues with the load.
Regs,
Jiri
are there any recommendations / experiences with high values for the
-mp parameter for the async collector? Somewhere in the code it says
10 to 20 are reasonable values. But now I've run tests with mp 10000
and see no issues so far.
We currently poll some 5000 targets on (mostly) Cisco routers and
switches. Some with 1min polls and some with 5min. This works fine,
ideally the collection is done in less than 30sec.
But if there is an outage in the network, many devices won't reply,
causing timeout delays in snmp collection. If you collect with -mp 10,
you risk to have a lot of these timeouts in sequence, and the
collector won't finish before the next polling cycle.
If you have mp higher than the actual number of targets, all async
requests are sent out right away, and there is only one waiting for
timeouts in the end.
Are there any long-term experiences with such high mp settings? I have
it running, but no big network outages yet. And the collection server
(Linux x86) doesn't seem to have any issues with the load.
Regs,
Jiri