Quote Originally Posted by SeanLeyne
Joost,

Given that the "spike" represents an increase to little more 1% CPU utilitilization, I think you are reading way too much into it. Think of it as a comatose patient who every so often twitches to show that there is some brain activity taking place...
Hi,

where do you take the 1% from? If I'm not lacking coffee, the diagram shows load not CPU utilization.

So the sporadic increase to above 1.0 means that one CPU (more precise: core) is fully loaded. If this were a single-core machine, this would be rather important.

As to the source of the spikes: I don't have my load graphics at hand right now, but don't recall seeing these on a DSS server without client load. OTOH, I do see such spikes whenever some client accesses ie NFS volumes. So when you have periodic client access (ie backup software scanning for modified files, batch processing or alike) I'd expect to see such spikes.

You might want to compare against the CPU utilization chart to see whether the load is caused by actively doing something on the DSS server (increase in CPU util - user or system space - which could be the handling of iSCSI packets) or by I/O (I/O waits growing during the load peaks). Also you can compare to the network I/O - if you see traffic increase between server and client, then obviously those are causing the load. If it's just traffic between the servers, then it could be some replication action.

Regards,
Jens

PS: May be bean counting, but II don't see *hourly* spikes - it's more like 45 minutes. This may be important in identifying the periodic processes.