[MERGE] Repository.get_revision_graph improvements

[Robert Collins]

On Wed, 2008-02-27 at 09:30 -0600, John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Robert Collins wrote:
> | On Tue, 2008-02-26 at 17:30 -0600, John Arbash Meinel wrote:
> |> If we instead chose a fixed hash algorithm to say 32-bit integers then
> |> the combined index can have 1 single hash => (index, offset) map. So
> |> when looking up revision-foo-bar you hash it 1 time, and get back the
> |> possible indexes and positions right away.
> |
> | I think there are two problems in this proposal:
> | a) the in memory hashtable becomes very big. This is a scaling problem
> | itself.
> | b) it will force pack-count reads to start any operations, which is also
> | a scaling problem (rather than being a missed-lookup problem)
> |
> | That said I don't see my design prohibiting combining of hashtables as
> | long as the same hash is chosen before the truncation.
> |
> | -Rob
> |
> 
> 
> Actually, you don't have to have it fully populated all the time. Just have it
> "these are the hash entries I've seen". If you get a miss, then you go load the
> next index in the list.

well, perhaps. But still - loading the full hash table from an index is
very wasteful.

-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/20080229/79fe8de9/attachment.pgp