B+tree discussions

[Robert Collins]

On Thu, 2008-07-03 at 18:33 +1000, Robert Collins wrote:
> At this point, I think we have a 'good enough to be an improvement'
> candidate, once some remote testing is done. (And I'm on that now).
> 
> B+Tree, no bloom, with 1000 page cache outperforms or is equal with
> GraphIndex on every benchmark I've come up with. I think in real world
> larger-still indices it should be tolerable if not better due to lower
> memory consumption, and it should be tunable (simply by upping the cache
> size) to deal with such cache thrashing workloads.
> 
> I think bloom use is a compelling future feature, but its not a clear
> enough win /today/ and we can take the time to make it better later.
> 
> If everyone agrees with this concept, I'll make the stackable formats
> about to be merged, also include this index layer.

So, I will move on this next week, unless someone beats me to it.

My plan is (in case someone wants to jump in):
copy the btree_index.py, tests, _parse_btree_c and _py files,
chunk_writer to bzrlib verbatim.

The repofmt.py file should be merged into pack_repo.py, and altered so
that the bulk-copy stuff is instead just changed in the base class and
parameters given to constructors.

-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/20080704/7a795e7a/attachment.pgp