Am Mittwoch, den 13.12.2017, 17:08 +0000 schrieb Ken Moffat: > On Wed, Dec 13, 2017 at 10:07:07AM -0600, Bruce Dubbs wrote: > > Thomas Trepl wrote: > > > Am 2017-12-13 13:39, schrieb Thomas Trepl: > > > > hi all, > > > > > > > > I'm currently trying to compile version 5.10 but it seems my > > > > machine > > > > is too small. It is a VBox-VM with 2GB RAM, 2GB swap partition > > > > but > > > > compilation failed with "virtual memory exhausted". I increased > > > > the > > > > mem to 3 GB but it looks like the process runs in same > > > > situation > > > > (currently at 13957/21739 targets to compile). > > > > > > I think such a small system will have pains with several large > desktop packages. I haven't built in a VM in ages, but on my i3 > (4GB RAM, 3.75GB available) and my A10 Kaveri (8GB RAM but < 7GB > available) building qtwebengine is unpleasant. Yes, its not that highly equipped. Looks like I have to tweak the VM settings a bit. I just wanted to hear your opinions whether tis config is indeed to small and such monsters of packages to require a better equipped machine. > > How do the 4 CPUs compare to what your host system offers ? My > impression has always been that *qemu* will not map CPUs 1:1 (there > is an overhead of running a VM) and for a big package that DOES make > a difference. But I have no idea about vmware. The physical box is a 16GB E3-1245 Xeon, showing 8 CPUs to the system. Sure, there is a bit of difference from VM compared to physical, but the system at all is that fast that I do not worry about.
> > Also: qtwebengine, at least until 5.9, has used its included ninja > if there is no system version. And ninja spawns N+2 jobs if more > than 2 CPUs. For system ninja we have Bruce's patch to use an ENVVAR > to limit the number of jobs. Limiting the number of jobs might help > in a VM or anywhere else short of memory - these are not trivial > 'build some little C file' jobs, these are "throw a kitchen-sink of > headers at this complex C++ code" jobs. Ah yes, the note in the BLFS book that ninja will use all available cpus made me forget that Bruce has created that patch to address that. Will try it - for this package it seems to make really sense. @Bruce, thanks for the patch! > If you only have 4 (real) cores, using -j6 on large C++ files is > unlikely to help - it might hinder, it might not make a lot of > difference. Well, i configured my VM to have 4 CPUs as many of the BLFS packages has a SBU number with a note like "XX SBU (using 4 cores)". So i made 4 CPUs visible. > For a clean build, ninja has few benefits. In cmake packages it is > marginally faster than the Makefiles cmake produces, but only a > little. The benefit is for developers who (re) build in a dirty > tree after making a minor change. In that situation Makefiles > created by cmake can do a lot of work, good Makefiles from > autotools, and ninja files, will just check what needs to be > rebuilt. Thanks Ken and Bruce for the comments so far! -- http://lists.linuxfromscratch.org/listinfo/blfs-dev FAQ: http://www.linuxfromscratch.org/blfs/faq.html Unsubscribe: See the above information page
