Everytime the modeling occurs (every 12h), I see event queues and the zenperfsnmp process completely stops monitoring devices (see pictures below). Looking at the the logfiles confirmed the zenpersnmp process is not doing anything during modeling. As the modeling lasts about 1h, the monitoring is not reliable enough.
my settings:
- zenoss core 3.2.1
- ubuntu 10.04 LTS
- virtualized on esxi 5.1
- 4vCPU
- 12GB of RAM
- 3GB of swap
- 150 devices monitored, ~50% cisco networking device 50% microsoft servers
I've already done the performance tuning according to as described here http://community.zenoss.org/docs/DOC-10163.
- 103'000 objects in zope DB
- data.fs file is 500MB big
- zope.conf content:
...
python-check-interval 1133
...
<zodb_db main>
mount-point /
# ZODB cache, in number of objects
cache-size 200000
<zeoclient>
server localhost:8100
storage 1
name zeostorage
var $INSTANCE/var
# ZEO client cache, in bytes
cache-size 1024MB
# Uncomment to have a persistent disk cache
#client zeo1
</zeoclient>
</zodb_db>
...
- my.cnf content:
innodb_buffer_pool_size = 1024M
innodb_additional_mem_pool_size = 32M
What shall I try next ? the tuning about didn't help. the server only uses 20% of RAM. no swapping. and still the same problem.
thanks in advance,
lionel