From: Beman Dawes (bdawes_at_[hidden])
Date: 2003-09-21 07:53:53
At 03:23 PM 9/20/2003, David Abrahams wrote:
>For those people running the Boost regression tests, I suggest you try
>invoking bjam with -jN to do some of the build jobs in parallel. Even
>on uniprocessor systems it seems to help quite a bit to compile/link
>while some other process is waiting for the disk, and on
>multiprocessors I imagine it would be even better. I am using -j4 on
>my uniprocessor laptop now, for example.
I tried -jN a while ago when I got a hyperthreading CPU, and it did seem to
improve build speeds as long as GCC wasn't one of the compilers.
I didn't spend a lot of time diagnosing the problem with GCC, but what
seemed to have been happening was that GCC used so much memory that running
multiple compiles chewed up all the machine's RAM and more. That kills file
caching, which in turn kills performance.
Did you try with GCC? The problem was particularly acute with the old
version of the random library - now that it has been updated the problem
may have disappeared.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk