Are 'average' data counter values available from the vSphere API at collection levels 3 and 4 computed precisely (i.e., by summing up values and dividing either by a count or by a time period) or is there some sort of sampling involved?
The difference is important if one is trying to recover the original sum for the particular period of time. If the average is precise, then we should be able to obtain the original sum by multiplying the average value by the count or the time period, depending on the statistic. However, with sampling, the result of that multiplication may underestimate or exaggerate the original sum.
For example, net.received.average and net.transmitted.average give us average kiloBytesPerSecond over a known time period. In this particular case, the question is: are we going to get the accurate count of bytes received and transmitted during the period by multiplying the averages by the length of the period (300 seconds or whatever)?
Perhaps the answer depends on the counter type. I am primarily interested in network bytes, disk bytes, and disk latency statistics.