[Beowulf] Time measurement

[Greg Lindahl]

>2) include start_time , end_time code in the program : this won't
>include the communication , i/o , loading etc overhead.

By the way, many parallel benchmarks do it this way. Why? Because you
may have to run for a long time to have a typical ratio of startup or
initialization to real computation -- many apps are run for hours
after a minute or two of startup.

By measuring only the computation, you can do it much quicker, minutes
instead of hours. The only bad aspect to this is that you may not
realize that a given system does poorly on startup.

-- greg