On Thu, Feb 23, 2012 at 4:00 PM, Grant <[email protected]> wrote: >>> > Parallel builds are not deterministic so if the Makefile allows a race >>> > condition to develop it's pot luck whether you'll be hit with it or >>> > not >>> >>> I got sick of stuff like that so I run MAKEOPTS="-j1" on all of my >>> systems. >> >> If it were a frequent occurrence, there may be some benefit in that. But >> using only one of the CPUs 8 cores is such a waste when this sort of >> thing happens only every few weeks. Usually trying again works, rarely >> does using -j1 make a difference and when it does a bug report ensures >> that it won't be an issue in future. > > OK you've inspired me to give it another try. So if I find a package > that doesn't build with -jn where n > 1 but does build with -j1 I > should file a bug?
Pretty much. It can get more specific than that, but that much is already a help. Here's the relevant portions of my MAKEOPTS and EMERGE_DEFAULT_OPTS which should speed things up for you about as much as possible. MAKEOPTS="--jobs --load $n" # Where $n is num_CPUs * 1.25 EMERGE_DEFAULT_OPTS="--jobs --load-average=$m" # Where $m is num_CPUs * 1.5 With the "--jobs" parameters, I haven't needed to set $n or $m to num_CPUS*2 to try to keep the load average up. Here's my MAKEOPTS and EMERGE_DEFAULT_OPTS verbatim, for an eight-core machine: MAKEOPTS="--jobs --load 10" EMERGE_DEFAULT_OPTS="--jobs --load-average=12 --verbose --tree --with-bdeps=y --keep-going" If you want to keep things simple, just go with num_CPUs=n for both $m and $n. -- :wq
