On Sun, Mar 02, 2014 at 04:05:27PM +0000, Richard W.M. Jones wrote:
> On Sat, Mar 01, 2014 at 02:58:41PM +0530, Kashyap Chamarthy wrote:
[. . .]
> >
> > A single `make` job timing to compile everything on a systemd-nspawn:
> >
> > real 31m9.792s
> > user 17m18.359s
> > sys 13m17.868s
> >
> > For comparison, on the _host_, the same single `make` job timing:
> >
> > real 13m41.440s
> > user 13m5.816s
> > sys 1m9.911s
>
> These results don't make much sense to me. I would expect make to
> take a similar time on both.
Hmm, I didn't start out with comparisions as I was testing it for the
first time. I'll conduct another test (more carefully, this time) with
newest systemd from Rawhide and see how it fares. (Also, will try with a
libvirt-lxc and report how it goes.)
>
> Do you have a proxy/cache that could be caching the RPMs that
> yumdownloader fetches during the build?
I don't have a proxy/cache (should really make one!). However the Fedora
mirror is on the same network of the machine where I the test on.
>
> Did you do the second build in the same directory as the first build?
No, it's done in my home dir -- ~/kashyapc/src/libguestfs. First one is
in /src/test-container
> 'make clean' intentionally doesn't clean up some things like the
> appliance and test images which require lengthy rebuilds. You have to
> use 'make maintainer-clean' or 'make distclean' instead.
>
> libguestfs make (and especially 'make check') is a good stress-test,
> but it's too complicated and irreproducible to be a good benchmark.
Yeah, trying to report benchmarks is always tricky.
Answering drago01's questions from the other email:
- Order: first I ran the compile test via nspawn container; next, run
the same test on the in my home dir (on the host)
- No, I didn't reboot in between
- Haven't cleared caches (I presume you meant to free the pagecache
-- $ echo 3 > /proc/sys/vm/drop_caches)
--
/kashyap
--
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxxx
https://admin.fedoraproject.org/mailman/listinfo/devel
Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
