BTree + CHK Inefficiencies

[John Arbash Meinel]

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


...
>   iii) Only cache the zlib compressed content. This is closer to 35
>        bytes/record. Which is pretty optimal, but means any key that we
>        actually want to read we still have to zlib.decompress().
>        (TIMEIT says it takes 88us to decompress a 4k leaf node, if we
>        did this for every key, that would be pretty bad. Maybe we could
>        do "if I decompress 2x, then cache"...)


Note that _parse_leaf_lines to turn the decompressed string into actual
StaticTuple(string) objects is 111us per page. So zlib.decompress is a
fair portion of the overall time. And while 200us is pretty fast, when
you have 7k btree pages, that is still 1.4s total time.

John
=:->

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

iEYEARECAAYFAkxYVvsACgkQJdeBCYSNAAPVfACfck0us2BnS+8Ut/SdwqLzNMXx
PKQAmwbBGMLCFBogr5pJdNCvJD91OiRb
=KQwf
-----END PGP SIGNATURE-----