Parallel processes interfere with each other in many ways, by competing for shared resources such as shared caches, memory, disks, etc.

Would it be possible to determine latent factors just with a runtime analysis, by means of comparing the runtime of the respective applications while running them many times in different combinations? Or would the required sample size be too big for the method to be feasible?

If comparing only the runtimes would not result in significant results, which variable would be best suited to observe? L1-cache-misses, L2-cache-misses, page faults, total memory usage, tlb-misses, etc.

  • $\begingroup$ Usually, you measure the impact of parameters by fixing all others, and varying only one. Also, what is your platform? You can't measure all the things you list on every platform. $\endgroup$ – Raphael Oct 17 '12 at 20:10
  • $\begingroup$ The linux perf tools are capable of measuring these factors. $\endgroup$ – nairboon Oct 17 '12 at 20:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.