Accelerating push via bzr:// and friends

[John Arbash Meinel]

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

Aaron Bentley wrote:
| Robert Collins wrote:
|> On Tue, 2008-05-20 at 08:39 -0400, Aaron Bentley wrote:
|>> -----BEGIN PGP SIGNED MESSAGE-----
|>> Hash: SHA1
|>>
|>> Andrew Bennetts wrote:
|>>> Initial tests suggest that bzr.dev has regressed a bit since 1.4!
|>> I noticed that on Friday.  A lot of these get_parent_map calls are being
|>> done for heads, and John's recently done a lot of work to make heads
|>> more thorough and accurate, so that may be the cause.
|> This seems likely - its why I emphasised doing one and one one query per
|> loop as an important goal in the graph code (because while pack indices
|> may be a touch slow, network latency is even higher).
|
| Well, there are several things.  One is that heads is going really
| deep-- when I was pushing a branch of Launchpad, it got back into the
| arch-imported history.  So we need to make queries scale up in size to
| cope with repetition, and ideally we would change heads to not go so deep.
|
| Aaron

Well, we can double check why this is happening, but I have the feeling we are
encountering shortcuts that require deep history to resolve.

I understand Robert's discussion about doing 1 request per loop. It is something
I did in the past, but didn't find a big improvement. Certainly with network ops
we are likely to see more of an impact.

I'll look into it a bit.

By the way, Aaron, do you have these branches with the specific histories available?

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkgzHGwACgkQJdeBCYSNAANP2gCgmWPqndA5b9yP1PNg6USm1j3S
36cAnRS9bxHss3Z2R74JYr1oZCbujO4W
=llu4
-----END PGP SIGNATURE-----