I am simulating a system where the clients' unavailability is one big issue to the systems behavior.

I gathered the real system logs to analyze the unavailability of the clients in the deployment environments and I computed the values for Mean Time Between Failure (MTBF) and Mean Time To Repair (MTTR).

Now I want to simulate such a system failure in my simulator. Currently what I am doing is as follows:

1. Every client rolls the dice every 1 minute (Uniform random in [0, 1])
2. If the number if larger than a specified probability (P_unavail),
   the client becomes unavailable for the next (T_unavail) minutes (i.e. (T_unavail) = 10min).

The simulation does not need to capture the whole details of the real world. Showing the trends (how often the failure occurs? how long does the repair take?) is indeed enough. However, the simulation method and parameters should not be arbitrary.

If any one who has some knowledge on it.

  • Is the method that I am using appropriate?
  • If yes, what values for P_unavail and T_unavail would be appropriate? Can I compute them MTBF and MTTR from the log analysis?
  • If no, could you give me any idea?



1 Answer 1


Fix some probability distribution for failures and failure duration and sample from there. Maybe a Poisson distribution for failures and a normal distribution for repair times would make sense. Compare the results from your sampling with the real data to see whether things match.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.