suggest 'make -j2' for SMP machines?
georgeb at linuxfromscratch.org
Mon Jun 4 20:16:58 PDT 2007
Deskin Miller wrote:
> On 6/4/07, Miguel Bazdresch <lfs-01 at thewizardstower.org> wrote:
>> Deskin Miller wrote:
>>> [ should the book say anything about 'make -j X' on multi-core systems? ]
>> ['make -j X' where X is number of cores is more or less optimal...]
There are issues when running make with other than -j1. Some package
builds will fail when make tries a parallel build. You can check out
jhalfs' optimize section which has an incomplete black list of packages
with MAKEFLAG issues.
>> As far as mentioning it in the book... I'm not enthusiastic, since
>> it's basic knowledge of how your computer works... OTOH we already
>> say some pretty basic things, and the proliferation of multicore
>> CPUs might warrant a mention of make -j.
This also applies to the 'older' Hyper-Threading Intel products. I
saw a noticeable reduction in compile times when I went to -j2.
> The concept of matching 'make -j X' to the number of cores isn't
> difficult, but I'd argue it's worth a mention as much for those who
> haven't had cause to wonder about 'make's various flags before, as to
> give people a lesson in multi-core theory: in my own case, as a Linux
> user for several years now, make -j was an eye-opener when I came
> across it only yesterday, since I had never had a SMP or multi-core
> system until very recently upon which to compile, and thus unwittingly
> filtered out from my Linux knowledge the make flags which didn't have
> any use to me, until now.
> In short: say it for those who lack 'make' knowledge, as well as those
> who lack 'multi-core' knowledge.
> -Deskin Miller
More information about the lfs-dev