btree - blooms

[John Arbash Meinel]

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

Robert Collins wrote:
| John - what are your thoughts on btree inclusion in core?
|
| As I see it we can:
|  - include bloom support
|    - disabled
|    - enabled
|  - not include bloom support
|
| I don't have a real preference at this point, though I think including
| blooms will require us to do some C for performance.
|
| -Rob

I've done a reasonable amount of pyrex work to have blooms functional at
a speed that meant they were helpful (locally). There are still
questions about paging (at the moment, it always loads the full bloom).

I think it would be very prudent to test performance against a GC
repository, rather than going off the benchmarks for a btree pack repo.
I would like to both test locally and over something with latency, to
see if blooms help more in that situation.

I'm a bit concerned that always loading the full bloom will impact the
"I just need 3 keys" sort of performance when you have a large pack. I
don't think it is hard to implement a paged bloom, I just didn't get
that far.

So... I would like to do a bit more testing on it before giving a final
decision. My results to date show a minor improvement on local access
for certain actions.

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

iEYEARECAAYFAkiYUjoACgkQJdeBCYSNAAMcwgCfS2RMGOXK5zn2LVOiyHQI9H9v
I+AAoJWUhQJmSmxslhdDXGSespoq8mTe
=9lFs
-----END PGP SIGNATURE-----