A best practice for VMAN integrated with NPM is to match the polling intervals between the two environments. If you take the NPM defaults (10 minutes) this isn't problematic, but if you start tweaking your NPM stats collection down (say 5 minutes) then you start to introduce a difference in statistic granularity. In very large vCenter environments data collection can last a few minutes (not < 60 seconds as seen on smaller vCenter servers). Being able to accurately judge if that polled vCenter server is taking 4:01 or 4:59 to collect data lets you know if you have to engage support to do some tweaking on data collection performance or if you have some breathing room.
I know it is a little thing --- but the little things count!
