In a talk on RED method, the presenter defends his use of 'average' by the following comparison with latencies. The linked-to part of the talk written down (non-verbatim and with my formatting):

There's 2 reasons why I plot averages:

  1. Latencies don't sum. If you got
  • A) 1 service that has high latency and you look at the 2 services it calls. Often it happens that the latter 2 have low latency. And this doesn't help you. Because latencies don't sum like that.
  • B) Averages (on the other hand) do sum. So if you got 1 service with high average, then one of them below, one of the 2 mentioned services it calls, is gonna have a high average. And that helps you pinpoint.
  1. (2nd reason omitted.)

What does he mean by, if I may summarise it, 'latencies don't sum, while averages do'. Also, wouldn't we be able to immediately pinpoint the callee service as the problematic one (as it's the only one with a high latency)?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.