Early numbers on multi-parent diffs

[Aaron Bentley]

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

Aaron Bentley wrote:
>> One of the ones I'm most interested in would be inventory.knit. Care
>> to try it there? (I realize that will take the longest, but might also
>> see the largest gain).
> 
> Yeah, I had to disable caching to get it to complete in a reasonable
> timeframe.  For some strange reason, the time to retrieve a given
> version is slower for MultiParent than for standard knit: 10.2s vs 5.2s.
>  I'll have to investigate further, but my guess is that the snapshots
> are suboptimally placed, causing us to go far past 25 revisions
> following the righthand parent.

Yes, it turns out that ensuring that each revision is no farther than 25
revisions away improves the speed dramatically.  It takes 0.29 seconds,
rather than 10.2.  But it roughly doubles the number of snapshots.
Reducing the snapshot interval decreases the number of snapshots, but
increases the graph size, harming performance.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFGHT3v0F+nu1YWqI0RAlhuAJ4uxTRicYr540a+k7XVTJfilECpogCffas+
V6zLmDR/2N8hqW4vVTXTJKU=
=Cvxo
-----END PGP SIGNATURE-----