On Thu, Nov 06, 2008 at 01:03:48PM +0000, Eric Kow wrote:
> On Thu, Nov 06, 2008 at 23:42:15 +1100, Trent W. Buck wrote:
> > Just for my peace of mind, can someone assert that -j4 significantly
> > increases build time even on single-core uniprocessor systems?
So far (the important number is "elapsed"):
$ map time bash -c '{ make maintainer-clean && autoconf && ./configure && make
$0 darcs ;} &>/dev/null' -- '' -j2 -j4 -j8
508.16user 25.79system 10:27.42elapsed 85%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+60112outputs (0major+2373859minor)pagefaults 0swaps
514.53user 26.14system 10:20.72elapsed 87%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+60112outputs (0major+2374572minor)pagefaults 0swaps
520.84user 27.00system 10:13.96elapsed 89%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+60128outputs (0major+2374151minor)pagefaults 0swaps
The first three have all taken between ten and eleven minutes, with
slight time decreases and CPU utilization increases proportionally to
the jobbiness.
Note that these tests are done with a PCIe SSD rather than a
conventional SATA HDD -- I believe this makes my disk I/O considerably
slower than a conventional machine. My /tmp is a ramdisk, however.
> I'll be also interested to see this
>
>> Incidentally, why -j4 and not -j2 or -j8?
>
> One reason (for me) is that we have 3 sets of tests + 1 network test,
> which adds up to 4 things that can sensibly be run in parallel.
> Anything more would be overkill (?). I want to make sure the network
> test is kicked off in the beginning because it involves a lot of waiting
> around.
Perhaps we should change the test to
make -j4 disttest # -j4 to parallelize test-network & test-%
> I assume it is ok for me to push this patch
Agreed.
_______________________________________________
darcs-users mailing list
[email protected]
http://lists.osuosl.org/mailman/listinfo/darcs-users
