brian.wilkinson@??? said:
> Are there any guidelines as to what size a file needs to be to merit
> using cdb or dbm format? Does cdb format offer significant
> performance benefits over dbm?
For cdb, which is very lightweight, I would guess that a file that is
more than a couple of disk blocks in size would be faster in cdb format
than linear search. You could find pathological cases which would be
faster either way (and remember exim caches internally), however I'd
personally move to cdb at a couple of K total size or a hundred keys.
[Actually I tend to always start with cdb - you may find very few cases
where its slower than linear searches even with small numbers of
entries].
cdb is orders of magnitude faster than any of the dbm implementations I
tested against a few years back. Specifically at that time it was much
faster than berkeley db which was faster than the other
implementations. I doubt the others have got much faster especially as
berkeley db carries around locking code as well - cdb can optomise as
its a read only database.
Am I biased - you bet - I wrote the exim cdb code. However I did check
out the opponents before putting the effort into writing that code.
Nigel.
--
[ Nigel Metheringham Nigel.Metheringham@??? ]
[ Phone: +44 1423 850000 Fax +44 1423 858866 ]
[ - Comments in this message are my own and not ITO opinion/policy - ]
