[libvirt] [PATCH] dist: Speed up distribution compression

Daniel P. Berrange berrange at redhat.com
Fri Jul 1 08:54:10 UTC 2016


On Fri, Jul 01, 2016 at 09:18:34AM +0200, Ján Tomko wrote:
> On Thu, Jun 30, 2016 at 03:13:58PM +0100, Daniel P. Berrange wrote:
> > On Thu, Jun 30, 2016 at 04:10:01PM +0200, Martin Kletzander wrote:
> > > This almost reverts b7200d723648.  The size is increased from 11M to 13M
> > > and the compression is sped up from 2 minutes to 17 seconds.
> 
> At the expense of download and decompression time.
> 
> > > The
> > > compression level is removed because -9 doesn't allow multiple threads
> > > to be spawned.  Effectively speeds up distcheck as well.
> > 
> 
> 1) For distcheck, any level over 1 is a waste of time, since we're throwing
> it away.
> 
> 2) For regular libvirt distribution, any time spent compressing results in
> greater overall time savings, because the file is downloaded and
> decompressed many times.
> 
> So, should we keep moderate defaults here, with anyone wantintg to do
> 1) or 2) properly supplying their own XZ_OPT? Or create a different
> set of targets with lower compression settings for development purposes
> (e.g. quick-distcheck, quick-rpm).

I don't think it is worth the complexity of trying to do things differently
for development vs formal releases.

Remember that the previous 1.3.5 release of libvirt came in at
33 MB, and in reality no one has ever complained about the size
of that. So whether we get a new size of 11 vs 13 MB with xz
is really irrelevant IMHO - its already an incredibly huge win
over gz. So picking the compression level to prefer speed at the
cost of a small extra size is really the right thing todo.


Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|
|: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|




More information about the libvir-list mailing list