I fixed some package (with small patches). I will send tomorrow my proposed patches.
best regards, luigi On Wed, Oct 28, 2009 at 9:29 PM, Felix Fietkau <n...@openwrt.org> wrote: > Michael Buesch wrote: >> On Wednesday 28 October 2009 03:04:41 Felix Fietkau wrote: >>> > That would be a huge advantage and speedup the openwrt buildprocess by >>> > about 35-40 percent points on my quad machine. The advantage gets higher >>> > the more cores we have. >>> > And I think quads are not that uncommon anymore today. >>> Why do you think the difference is that big between building multiple >>> packages at the same time and building individual packages in parallel? >> >> Because the cpu utilization is horribly low during an openwrt build. >> It ranges from 25% for the toolchain to 99% for the kernel. >> For compiling "normal" packages there's usually an overall utilization of >> 60-75%. I'm pretty sure >> if we'd allow two processors (or more) on one package at the same time, it >> would go up to nearly 99% >> on my machine. Especially for bigger packages like busybox, ssh, etc... >> I'd also like to have an optional (default off) parallel build of the >> toolchain. I don't mind >> poking it two or three times. That'd still be faster than compiling the >> whole toolchain at 25%. > I did a test with all packages of trunk selected (/packages not > included), and 3 packages broke with very weird build failures. > This was with a small change to include/package.mk that sets > MAKE := $(MAKE) -j 8 > and some other patches to replace $(MAKE) all install calls with > multiple calls. > The three packages that broke were apex, iproute2, openssl. Also when > I've done similar tests in the past, some other packages also broke, > though only under rare, hard to trigger circumstances. > If somebody makes a patch that enables package build parallelization, I > definitely don't want to have this enabled by default whenever somebody > does make -j in the OpenWrt build system. > I think most people that use make -j would prefer to have something that > may be a little slower, but doesn't cause unexpected build failures > depending on the parallelization level. > > - Felix > -- Luigi 'Comio' Mantellini R&D - Software Industrie Dial Face S.p.A. Via Canzo, 4 20068 Peschiera Borromeo (MI), Italy Tel.: +39 02 5167 2813 Fax: +39 02 5167 2459 web: www.idf-hit.com mail: luigi.mantell...@idf-hit.com _______________________________________________ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
