Date: Sun, 2 Nov 1997 16:12:47 -0500 (EST) From: Chuck Robey <chuckr@Glue.umd.edu> To: Wolfram Schneider <wosch@cs.tu-berlin.de> Cc: "Jordan K. Hubbard" <jkh@time.cdrom.com>, smp@freebsd.org Subject: Re: Some SMP timing tests. Message-ID: <Pine.BSF.3.96.971102160833.1115E-100000@picnic.mat.net> In-Reply-To: <p1ipvojjjo3.fsf@panke.panke.de>
next in thread | previous in thread | raw e-mail | index | archive | help
On 2 Nov 1997, Wolfram Schneider wrote:
> "Jordan K. Hubbard" <jkh@time.cdrom.com> writes:
> > > If you have a Makefile with 5 targets (e.g. 4 *.c files, 1 manpage),
> > > make can only create 5 jobs at once. Not surprising ;-)
> >
> > A good point, but this doesn't quite explain the behavior during
> > compiling things like libc, which has many targets which could be
> > parallelized.
>
> The bottleneck is the SUBDIR target which is not parallelized.
>
> Try the patch and run `make -j<n> -I share/mk all' in /usr/src.
> The patch split the SUBDIR variable into 4 variables SUBDIR1,
> ... SUBDIR4 and runs 4 subdir targets simultaneously.
OK, I tried it. I started with no obj at all, did a make obj outside the
timing loop, just so that I knew it was completely clean. This goes for
both runs. The first one is before the patches were inserted, the second
one after. Both on a Tyan Titan II, 2 each PPro 166's e/w 512K cache
each, 64 Megs main memory. I have set the source on the first disk, the
obj on the second, and swap evenly distributed between the two. Here's
the results:
before patching->
/usr/bin/time -l make -j 12 buildworld:
5389.14 real 3229.98 user 2902.01 sys
8200 maximum resident set size
249 average shared memory size
306 average unshared data size
58 average unshared stack size
7096181 page reclaims
2930 page faults
0 swaps
34729 block input operations
209264 block output operations
0 messages sent
0 messages received
8 signals received
903820 voluntary context switches
430529 involuntary context switches
With patched Makefile and bsd.subdir.mk, same command:
2780.55 real 1696.39 user 1571.73 sys
6828 maximum resident set size
270 average shared memory size
350 average unshared data size
60 average unshared stack size
4079428 page reclaims
2607 page faults
0 swaps
10619 block input operations
130129 block output operations
0 messages sent
0 messages received
0 signals received
473969 voluntary context switches
227592 involuntary context switches
Pretty major difference? Seems to me ....
----------------------------+-----------------------------------------------
Chuck Robey | Interests include any kind of voice or data
chuckr@glue.umd.edu | communications topic, C programming, and Unix.
213 Lakeside Drive Apt T-1 |
Greenbelt, MD 20770 | I run Journey2 and picnic, both FreeBSD
(301) 220-2114 | version 3.0 current -- and great FUN!
----------------------------+-----------------------------------------------
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.3.96.971102160833.1115E-100000>