We have several ESX 3.5 clusters but are experiencing performance issues with one cluster in particular.
Memory consumption http://molcat.bl.uk/gwicks/esxmemory20100216.png
Why does the cluster memory dip by 50% every hour at 10 minutes and 25 minutes past the hour. We think this may be ESX doing some kind of memory reclaiming, but we think it's also causing the application to slow down, something we observed during our load testing.
There are 8 guests, 4CPU and 16GB Redhat 5 servers with Java apps (Lucene text search)
OK - that question got you all thinking... We need to cultivate some vmware esx guru's on SF!
Anyway, after more investigation (and a call to vmware) the "spikes" are indeed where esx will, in turn, host by host, test the guests responsiveness to CPU and RAM requirements on an hourly basis. It does not affect performance.
The slow downs in performance my application is having is due to not enough RAM being allocated to the guests and the server running in HA mode when it does not need to.