On Tue, Aug 19, 2008 at 9:37 AM, Bas van Dijk wrote:

On Tue, Aug 19, 2008 at 8:49 AM, Johan Tibell wrote:

...

benchpress is a micro-benchmark library that produces statistics such as min, mean, standard deviation, median, and max execution time. It also computes execution time percentiles.

Nice, I'm certainty going to use this. Thanks!

It would be even nicer if you could also measure the elapsed CPU time (using System.CPUTime.getCPUTime). Than the measurement is not influenced by time spent in other processes and IO time.

Thanks. I will ponder how to best expose this in the interface. An option would be to return a tuple of Stats records where the first component is the wall clock time and the second is the CPU time. I might have to run the action twice to avoid extra measuring overhead. For example, in startCpu <- getCPUTime startWall <- getCurrentTime action endWall <- getCurrentTime endCpu <- getCPUTime the measured time includes the two calls to getCurrentTime. Running the action twice will double the benchmark running time which might or might not matter. Cheers, Johan

On Tue, Aug 19, 2008 at 2:14 PM, Bas van Dijk wrote:

On Tue, Aug 19, 2008 at 10:54 AM, Johan Tibell wrote:

...

I might have to run the action twice to avoid extra measuring overhead.

I don't think it matters because the extra measuring overhead is constant over the measuring iterations.

I want the reported times to differ as little as possible from the actual running times. Any constant overhead skews the results upwards. It might matter is the action being benchmark has a very low run time. Maybe I misunderstood what you said. -- Johan