Better compression

[Robert Collins]

On Fri, 2008-07-25 at 22:05 -0500, John Arbash Meinel wrote:


> In testing, I get really good results. It also revealed the need for my
> other patch to stream Revisions and Signatures (because otherwise my
> screen fills up with requesting revision 1-by-1.)

We need to check that GenericRepoFetcher is never used with altered
serialisers I think, but otherwsise the other patch should be fine. My
work for reverse-topo will be ready my Monday.

> These are the times for branching bzrtools into a gc repository
> 
> Time	Stream Revisions	LRU Cache
> 37s	No			No
> 24s	Yes			No
> 20s	No			Yes
> 8s	Yes			Yes...

> Anyway, a 10MB cache gives us a 93% hit rate for bzr.dev texts, which
> seems to make it more than worthwhile. I would consider tweaking it a
> bit, to try to make the estimated size more closely line up with
> reality. But I do take into account the python overhead of string
> objects, and the overhead of storing lots of lines in a list (at least
> 4bytes for a pointer per line, and probably 20+ bytes per string for
> PyObject overhead.)
> 
> The LRU Cache is in my groupcompress branch, and the streaming revisions
> is up for review.

I would rather not have the LRU cache to be honest; its working around
asking for the wrong stream type; and will just bloat memory once that
is fixed.

-Rob
-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20080726/743c5f76/attachment.pgp