Low level storage update status

[John Arbash Meinel]

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

Ian Clatworthy wrote:
| Robert Collins wrote:
|> So I've been doing some experimentation on real-low level storage
|> changes, aided by John and with various other folk doing testing and
|> feedback.
|
|> I intend to keep plugging away at these formats over the next couple of
|> weeks. What would be useful for me is folk to test the gc plugin and
|> file bugs tagged groupcompress and btree if they have trouble; Running
|> usertest over a gc-plain repository would be interesting though right
|> now probably not terribly surprising :).
|
| Thanks for the status update.
|
| I have done some limited usertest runs using groupcompress and I'll do
| more once the conversion speed is improved. Right now, it completely
| dominates the time taken to run the tests, particularly for the big
| interesting repos with deep history.
|
| When we think the format is "stable" and the conversion is inserting
| things in the optimal order needed to minimise size, I can do a
| conversion of the big repos and keep those around for testing the
| rest of the operations, say.
|
| Ian C.
|
|

I have a "trivial" patch which introduces a LRUCache that greatly
improved the conversion time. I think Robert is working around that by
optimizing fetch to request texts in a more appropriate order, as well
as have the source repository responsible for extracting fulltexts
(rather than using an adapter and pulling them out of the target repo.)

Anyway, my original testing said it took 90min to convert bzr.dev, with
my last few patches it was down to 7 min, and Robert says his approach
has already gotten him down to 28min. (I don't know how our approaches
will stack with eachother yet.)

John
=:->

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

iEYEARECAAYFAkiR1N4ACgkQJdeBCYSNAAPr1gCePx97bqcMOOnhIjvXvXH+lXDt
uRUAn3Eatyb60H4nM5YqVfFWeRAsLASw
=3QrO
-----END PGP SIGNATURE-----