After some efforts, I'm really doubtful about the application of that queueing model in analyzing network servers (specially in my case, IP Multimedia Subsystem) as well as predicting the workload. Therefore, as a newbie in this area, the purpose of my question is to ask if somebody can give me some advice for my following issue.

I am performing some analysis to measure network metrics (i.e. delay) or resource utilization by using M/G/1 queueing model. But my issue is that the analytical value is too much different from the experimental value (with Poisson traffic generator). For CPU utilization, I realize that it varies time by time and really is not what I expected based on analytical model with different inputs (i.e. arrival rate). And same problem with delay.

I've seen several papers that adopt M/M/1 and its variants to model a system to measure the delay. The point is that many of them have validated their model via simulation instead of doing real experiments (via testbed or in production environment). While the others omit the queueing delay due to many factors.

Any idea or suggestion is highly appreciated.

  • $\begingroup$ Welcome to CS.SE! While interesting, this sounds like a pretty broad question to me (community votes?). I'm not sure if there will be any single answer that applies to all situations. Queuing theory will be applicable if its assumptions are valid. The theory makes a several simplifying assumptions; I suggest you try to build a list of those assumptions, and then go through them one by one and see if you can validate each one to see if it appears to hold for your practical system. That might help give you a lot more insight about what you should expect the applicability of the theory to be. $\endgroup$ – D.W. Sep 1 '16 at 22:14