Git and Linux tarball size evolution

http://bouillon.math.usu.ru/files/linux-tarball-evol.png

I plotted sizes of official linux kernel tarballs found at
ftp://ftp.kernel.org against their dates. (Yep, the methodology is
extremely dirty.)
Long story short: the normal dynamics of Linux development is
accelerated growth. Apparently, in the end of 2002, development hits
some limit, probably a scalability problem. It returns to the
accelerated mode in mid-2005, which event coincides with the
introduction of git (pointed by the arrow).
Speculation: indeed, the git lets development scale.

It is clear that git has changed the release pattern. But was it the reason why the development (and tarball size) returned to accelerated growth? Another possible interpretation is that 2.5->2.6 stage involved too much of reengineering, so "normal" incremental development slowed down for a while. Do git developers have any opinion on that? Thanks!

On Fri, Apr 9, 2010 at 18:33, Victor Grishchenko <victor.grishchenko@gmail.com> wrote:

Could you try using the sizes of *unpacked* tarballs? gzip/bzip2 will offset the real growth a bit (maybe even a big bit).

There were a lot of cleaning up in 2.5/2.6.

The was Bitkeeper before Git, but also the development process has changed, with Linus becoming less of nexus of it. He does more merges than ever now, with a large part of integration and testing done by subsystem maintainers and people like Andrew Morton. Besides, you cannot ignore the developments outside of Linux world, which percipitate into kernel (things like new architectures).