Lars,
I can tell you what I have done in my environment. I'm not saying that this is the best answer, but it works for us.
1. I have edited the alarm definition for Guest Memory Usage under datacenter, alarms, definition.
It is set to Warn if condition is above 85% for 15 minutes, or Alert if condition is above 95% for 10 minutes.
This way when a VM boots, it touches 100% of the memory, but over the first 10 minutes some of the memory is not used. As long as 5% isn't not used after 10 minutes, the Alert trigger will not hit. Then in the next 5 minutes, the amount of active memory should fall another 10% so that the warn flag should not hit. Now if either of these conditions do hit, and stay on, I know that I probably need to give that VM more memory, as it is actively using the memory I have given it.
Most VMs will have a sweet spot. I have a SQL server running Dynamics, that was given 4GB and was constantlly alerting. I boosted it to 6GB, and it runs better and no alerts. I'm sure the sweet spot was lower than 6GB, but if memory becomes an issue in the environment, vSphere has memory ballooning and memory compression techniques for pulling back unused memory.
Doug
PS. To help you set the tresholds, try opening the advanced performance tab for a VM. Switch it to memory and reboot the VM. You will see the active memory spike (Red line in my graphs). Then after a few minutes you will see it start to settle down. For me this occurs in about 10 minutes, thus the tresholds I set above.
