Performance improvements for bzr-2.4 on large trees

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


...
> Ok, I have taken a quick look and saw some improvements but not as much
> as I hoped.  Let's not jump to any conclusions, because my experience is
> based on a repo with a fair amount of binary bloat (versioned DLLs):

So the #1 thing is that the main improvements are when the tree you are
dealing with is really big. If you go back to my numbers, I had 2
tables. The 70k entry table showed massive improvements, the 7k entry
table showed minor ones.

And times for something <1k entries is mostly dominated by python+bzrlib
startup time.

#2, I didn't specifically improve stat/add/commit times. I'm not sure
about add, but commit already was trying to use the update-by-delta code
(that was the code that I cleaned up to support any delta, rather than
only supporting commit deltas).
So the places you'll see the most benefit are the ones I mentioned.
merge/pull/update/uncommit.
...
> Platform is Windows XP, 32 bits.  As close to zero load as I can get. 
> Just a "DOS" shell and python.
> All tests were performed with --no-plugins and I verified that I was
> running the intended revisions by checking 'bzr version'
> The machine is an old dual Opteron running at 2GHz with 2GB RAM.  Python
> is 2.6.5 and I get the usual warning about some compiled extensions not
> being loaded but I don't think that matters much for John's changes.

> Since I am not seeing as much improvement as John, I speculate that I
> might be limited by either memory or filesystem throughput.  I will try
> a machine manufactured in this decade soon.

There are a few possibilities, but primarily you're not running commands
that I improved.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk3VmIUACgkQJdeBCYSNAAMKAQCg0970PqgIYCra82+YNrk1S2dR
ansAnRM8pG4WrA7WkPAUjg6n/7FY3eZN
=vRFh
-----END PGP SIGNATURE-----