Exactly as described, for the latency and PDV tests specifically, this sets the number of samples taken in order to determine the minimum, average, and maximum latency, and average PDV. Because latency and jitter can vary sample to sample even on a healthy link, this value directly trades off statistical confidence for test time -- a low sample count risks an unlucky single spike skewing the reported maximum latency, while a higher count smooths that out at the cost of a longer test. For example, using only the minimum allowed 2 samples risks a single transient spike being reported as the maximum latency, while the maximum allowed 20 samples gives a far more statistically reliable picture of the link's real latency behavior.