[Date Prev][Date Next] [Chronological] [Thread] [Top]

RE: HDB Tuning



--On Thursday, January 18, 2007 9:32 AM +0900 Mark Mcdonald <mmcdonald@staff.iinet.net.au> wrote:

-----Original Message-----
From: Buchan Milne [mailto:bgmilne@staff.telkomsa.net]
Sent: Thursday, January 18, 2007 12:17 AM

Is there any reason you didn't include BDB as well ? 

We have a reasonably high write component in our system so we thought
we'd go straight for HDB.  From my fairly limited reading, I get the
impression that HDB >= BDB, although I've seen nothing criticising HDB
comparative to BDB.  Is it something we should also consider? For what
reasons?

BDB and HDB are generally equivalent in search rates, IIRC. I haven't done a direct comparison in a while though. However, I switched to using HDB on my servers and they've been quite responsive.


How many entries in the database, or how large is it (du *.bdb) ?

1.3GB, 800k entries in one database. No glue or other overlays.

Really, the important part is, how large is id2entry. You generally want your DB_CONFIG database cachesize to be id2entry + at least 10% for growth. For loading the database via slapadd, you want the size of *.bdb. So I generally just set it to the latter number.

I think you need (more?) idlcache, configured in slapd.conf inside the
database. And, if you don't have any, you also need a cachesize
configured (in the same place). See the man page for slapd-hdb (or
slapd-bdb). IIRC the
guideline for idlcache on hdb is approx three times the cachesize (which
you
will have to decide on).


Bingo!  Adding the idlcachesize made a world of difference, thanks!  I
didn't check that man page so never realised the index caching config
directive had changed.  I've followed the 3x rule and it's super-speedy
now.


Yeah, the man pages are important. :)

--Quanah


--
Quanah Gibson-Mount
Principal Software Developer
ITS/Shared Application Services
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html