Comments (2)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 powers-old-account commented Permalink

Would it be statistically correct to get a random sample of the transaction times across the time period? For example, record every 10th, 100th, 10,000th or whatever transaction and record its transaction speed? This way you could spread out a reasonable number of random samples across your time period. <div>&nbsp;</div> I don't know much about statistics but a friend of mine in college was a stats major and we told me several times it's amazing how few data points you typically have to sample to get a reasonably accurate estimate of the data set.

2 MarkWeatherill commented Permalink

Sampling can be used for calculating an indicative value of the percentile. However customers don't find it suitable for measuring against a formal SLA. <div>&nbsp;</div> A significant challenge with sampling is determining the sampling rate. Since different transactions will have unique transaction rates, it needs to be configured on a per-transaction basis. This represents a risk as a mis-configured value could either invalidate the calculation or provide no benefit at all.