On Mon, Oct 01, 2012 at 10:55:14AM +0100, Colin Guthrie wrote: > 'Twas brillig, and Zbigniew Jędrzejewski-Szmek at 01/10/12 09:42 did > gyre and gimble: > > On Sat, Sep 29, 2012 at 09:02:11PM +0100, Colin Guthrie wrote: > >> 'Twas brillig, and Kok, Auke-jan H at 28/09/12 20:09 did gyre and gimble: > >>> 1) people should fix 'make' to just allow `-j` without an argument > >>> (seriously, dude ;^) ) > >> > >> Going dangerously off-topic, but two points: > >> > >> If you're using -j I've always gone under the impression you want the > >> value to be #cores+1, not #cores. That way you keep your machine working > >> full tilt. > >> > >> But regardless, why not use "make -l" anyway? This way it's tied to > >> system load which is likely a more prudent method to decide whether or > >> not new make jobs are issued - e.g. if you happen to be running a make > >> process that is io-intensive, you likely want to run less of them, but > >> if you've got a couple jobs one of which is io intensive then some of > >> your cpu cores might be mostly idle... -l should allow you to max things > >> out better. > > > > Yeah, -l seems better. But than you still want a load of #cores+1, no? > > And -l requires an argument too, so it has exactly the same > > inconvinience as -j. > > I thought -l was based on the Load Average of the system which is > separate from #cores. So if I like keeping my LA below 1.5, I'd just use > -l 1.5. and this wouldn't really matter how many cores I have. That > said, I'm prepared to be wrong here as I've not really read up about it > much! :)
I would love to be proved wrong --- i.e. to learn that 'make -l' adapts to the number of cores by itself. But as I understand it, "load" is "the number of running or waiting to run process". If 'make -l' uses the same definition, than one would want #nrcores+1 at least. Zbyszek _______________________________________________ systemd-devel mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/systemd-devel
