Memory usage- VROPs vs Netmon

[dbutch1976]

Hello,

Our Exchange team are insisting that their servers are under provisioned and running out of memory and paging to disk causing poor performance.  I don't believe this because VROPs and vCenter performance monitors show memory usage around 12-20% with rare, brief spikes up to 80%.  CPU usage is also very low.

Our Netmon monitoring tool tells a very different story.  It shows memory usage average 80%.

I'm trying to understand which of these statistics are accurate, obviously they can't both be correct and you can imagine which statistic the Exchange team believes is correct.  Where does VMware pull its memory usage statistics from?  Can these be relied upon?

In addition, because these are Exchange servers I believe they take all the memory they are allocated immediately (similar to SQL servers) making determining the real memory requirement a little more difficult.  Are there any guidelines surrounding this?  Are there better counters I could be using to get a real idea of the memory usage on these boxes?

Thanks.

[pratt26 · Accepted solution]

‌Exchange and SQL are more tricky to right-size for just the reasons you mentioned. Look at the Guest Demand Memory KB counter in VROPS (change the units to GB) for your VMs and change the timeline to 30-60 days. This will reflect the true memory demand and any peaks (which also need to be considered).

[dbutch1976]

I'll take a look at that, thanks!