Windows agent generates huge amounts of pagefaults

[Tom]

Hi,

I've noticed that the zabbix windows agent generates huge amounts of pagefaults on our Windows 2003 servers. More than 1000 pagefaults per second which i don't think is normal anymore.

Also it is swapping even tough there's only 50 % of physical memory used.

Any recommendations ?

[jacob.hamacher]

Zabbix agent generating page faults

We also observe high number of page faults on some of our win2003 servers. We have the same version installed on all servers, but the page faults only occur on less than half the servers. Our guess is that it has something to do with performance monitor counters.

On zabbix agent 1.4.2 with only standard counters we do not see these extreme number of page faults. But if we enable monitoring of SQL Server 2000 or IIS 6.0 we immediately get the high numbers. Removal of the offending items does not remedy the situation until after restart of the Zabbix agent service. The page faults does not occur at the same time an item is monitored (eg every 60 seconds for that kind of counter). The page faults run all the time. And it does not happen on all our servers, only maybe 30%.

My guess is that the Zabbix agent gets confused with some external performance monitor dll:s. Maybe it is loading and unloading dll:s in the main loop or something like that?

This is a dealbraker for us. I really hope to see this bug fixed. It has given the agent a bad reputation here, and we do not dare to roll out zabbix on more than a few of our production servers.

Are there any other that has seen this problem?


Regards,
Jacob Hamacher

[swaterhouse]

I have had lots of issues with the agent on Windows both with event log monitoring and performance counters. I never had any of these issues running the 1.1 branch - they all started after I migrated to 1.4

Mostly what I see are large cpu utilization spikes during event log checks and errors from perfomance counters.

I have to say that I really like ZABBIX but issues like this have made me think about evaluting other products.

[Tom]

I have to say that I really like ZABBIX but issues like this have made me think about evaluting other products.

Well i have evaluated all other products before i tried zabbix and i can guarantee you that you won't find anything better

Alexei, if you need any more debug information just tell me, while i do not have much room for experimenting because they are heavy production systems i can at least try to get some more information.

[RohrbaGe]

WIndows Agent poblems

How do you get the number of page faults, and are aware of it?

I have also some problems on a W2K machine, where sometimes
the agent 1.4.2 stops working without any findable reason.
Even with the debug switched on I can not find, why that happens.

What I noticed with netstat, there are a lots of connection to the zabbix server.


Gerald

[Alexei]

That's really interesting! What counter of Windows performance monitor can be used for monitoring of page faults?

[rolandsym]

windows counters

typeperf "\Process(zabbix_agentd)\Page Faults/sec" Specific to zabbix
typeperf "\Memory\Page Faults/sec" Overall but also includes faults that don't read from disk
typeperf "\Memory\Page Reads/sec" Mainly includes faults that read from disk.

I looked at this at it primarily looks like something that is affected by performance counters not zabbix. MS performance counters are creating these page faults. Which if "Page Reads or Writes" or "Pages/sec" had high numbers than there would be an issue which are the true measure of disk paging. Just my two cents.

Rolandsym