On Thu, 27 Nov 2008 13:47, Norman Ramsey wrote: > I wonder if anybody on this list has come up with any good tricks > for exploiting parallelism using ksh. Example: all the unit and system > tests I run on the code my students submit for homework are written using > ksh. Given that I have a four-processor machine, wouldn't it be > nice to get through those tests four times faster? But I'm not sure > I want to blithely launch as many processes as I have tests. > > Some form of work-crew parallelism would be ideal here. > The shortest path I have thought of would be to have ksh generate a > makefile that says what I want to have done, then use make to > limit the amount of parallelism going on at once. (For example, I might > want to tie up just 3 of my 4 cores running tests.) > > This procedure strikes me as a bit tedious. Does anybody have other > ideas? >
I can't see CPU being the limiting factor here; more likely I/O if the scripts are doing the usual business of typical shell scripts rather than a lot of string and/or maths business without much I/O. If that was the case, I'd be aiming to keep my disk queues nearly occupied without having too many requests stacked up; such a measure is far more nebulous. Some sar option shows the length of the physical disk queues and wait time averages to service it, but I can't remember. The buffers will also have some serious effect on the best balance too. All in all, I doubt there's a 1:1 correspondence between CPUs and optimum number of test instances, although I don't mind being proven wrong! _______________________________________________ ast-users mailing list [email protected] https://mailman.research.att.com/mailman/listinfo/ast-users
