Michael Brennen wrote:
> I was curious about this, so I ran a few quick tests this morning.
>
> Building wirehub as a hash file gave a file size of 5267456; the
> btree file size was 3411968. In one run building the hashfile took
> 0:19.85elapsed, btree 0:05.70elapsed.

Not quite on-topic, but I was curious too and compared berkeley-db
with cdb. Well, quite expectable results: cdb was *much* faster than
berkeley-db 3.2.9 in all modes. This wirehub's data allowed me
to do this (quick) comparision. Here are results:

Wirehub! spamlist-extended.txt, 2021461 bytes.

Create:
type real user sys size
hash 5.601s 5.340s 0.070s 5263360
btree 0.957s 0.680s 0.060s 3424256
cdb 0.156s 0.100s 0.060s 3504039

Query (method described by Victor):
type real user sys
hash 7.335s 4.570s 2.650s
btree 0.962s 0.920s 0.030s
cdb 0.120s 0.110s 0.010s

BTW, there is NO cache in cdb at all... ;)
(but mmap of the whole file is used - well,
optional).

It's interesting why query performance of
hash is worse than creation performance.

/mjt

-
To unsubscribe, send mail to majordomopostfix.org with content
(not subject): unsubscribe postfix-users

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]