Results of profiling parallel tests (was Re: [1194] grep in scalar context doesn't short-circuit (it can' t - side effects) so the)

[Nicholas Clark]

On Sat, Aug 30, 2008 at 02:57:17PM -0700, Eric Wilhelm wrote:
> # from Nicholas Clark
> # on Saturday 30 August 2008 14:48:
> 
> >ie time spent in the gubbins of the parallel test job scheduler has
> > dropped from "lots" to "bugger all".
> 
> ...and there was much rejoicing.
> 
> Where does that put you on the cpu vs wallclock compared to the serial 
> run?

parallel, TEST_JOBS=3:

Files=1553, Tests=202384, 222 wallclock secs (33.37 usr 11.58 sys + 251.99 cusr 49.48 csys = 346.42 CPU)
Result: FAIL

real    3m42.540s
user    4m45.621s
sys     1m1.100s


serial:

Files=1553, Tests=202384, 545 wallclock secs (31.38 usr  9.62 sys + 273.33 cusr 51.77 csys = 366.10 CPU)
Result: FAIL
Failed 3/1553 test programs. 3/202384 subtests failed.

real    9m4.804s
user    5m4.962s
sys     1m1.437s


(FAIL is the 3 Module::Build tests)


Um, so I'm actually using less CPU running the tests in parallel?
If so, that seems rather good.

Nicholas Clark