Date: Mon, 16 Oct 2006 16:21:18 -0700 From: Kent Stewart <kstewart@owt.com> To: Garance A Drosehn <gad@freebsd.org> Cc: freebsd-stable@freebsd.org, Nikolas Britton <nikolas.britton@gmail.com> Subject: Re: FreeBSD and "make -j# buildworld" usability Message-ID: <200610161621.18552.kstewart@owt.com> In-Reply-To: <p06240813c159b778bb5f@[128.113.24.47]> References: <20061013143130.GW491@dev.null.cz> <200610131347.41023.kstewart@owt.com> <p06240813c159b778bb5f@[128.113.24.47]>
next in thread | previous in thread | raw e-mail | index | archive | help
On Monday 16 October 2006 15:33, Garance A Drosehn wrote: > At 1:47 PM -0700 10/13/06, Kent Stewart wrote: > >On Friday 13 October 2006 07:31, Buki wrote: > >> Hi, > >> > >> I searched the archives and web a little but found many different > >> opinions on stability/usability of using make -j# with buildworld > >> (and buildkernel). > >> > >> So I am asking if it is a good idea to use make -j on production > >> boxes. > > > >I tested buildworlds with different values for -j. On single > > processors, using a script that basically looked like > > > >time make -j? ... > > > >yielded fastest builds when I didn't specify a value for -j. On dual > >cpu's a value around -j8 yielded the fastest build. > > You might want to repeat those performance tests. The results will > change over time, as various system changes are made. I used to see > results somewhat similar to yours, but the last time I checked I > seemed to get the best results from a dual-cpu system when using -j3 > . > > I believe that change in performance is due to some improvements > made to the `make' command itself, which solved some delays that > happened when -j was used. However, I do not remember the details. > > Performance comparisons like this will also depend on all the > hardware on the system, and not just the dual-CPU. It can change > depending on much memory the machine has, for instance, or how fast > the disks are compared to the CPU's. That was why I commented about making your own tests. I had ata-133 or 100 HDs on individual controllers. Comparing an ata based system with a scsi isn't a real comparison. My tests are at http://users.owt.com/kstewart/freebsd/urban_legends.html and I see I got the shortest elapsed time with -j4 and -j8. It has been too long to remember what I did to improve the -j8. They were run on FreeBSD 4.3 and I no longer have a dual cpu system. Kent -- Kent Stewart Richland, WA http://www.soyandina.com/ "I am Andean project". http://users.owt.com/kstewart/index.html
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200610161621.18552.kstewart>