[Date Prev][Date Next]
Re: 8 hours tests ends with inconsistent DB.
--On Saturday, June 12, 2004 7:31 PM +0200 paul k <firstname.lastname@example.org>
While totally agreeing with your deployment strategies I would not count
them as proper arguments for the matter at hand (benchmarking OL). I may
be ignorant but I'd expect a poorly configured system to perform poorly
and of course not doing things as you expect but not to misbehave or
fail. If the underlying DB is not configured explicitely, fall back to
sane defaults, if you hit the resource limits of hardware or whatever
mechanism, the application should behave gracefully.
I'm sorry, but what's being done here is not benchmarking.
If you are benchmarking a product, you
(a) Either make the data you are using available, or a description of it
available. That has not been done.
(b) Describe the configuration you are running under. Although this has
been done in part, it has never been answered fully, despite repeated
Things to think about:
50 million entries with 1 attribute are a lot different than
50 million entries with 70 attributes each.
Indexing 70 attributes is different than indexing 1 attribute.
Indexing all attributes as "eq" is different than indexing with a variety
of options, like sub and approx.
As for you question about sane defaults on the DB backend:
BDB is created and maintained by Sleepycat. *They* set what they think are
the sane defaults, not OpenLDAP. And, their default of 512kb cache may be
quite reasonable for some applications. It certainly isn't for OpenLDAP,
but OpenLDAP is definitely not the only consumer of BDB. At many levels,
the configurations in slapd.conf (cachesize, etc), entirely depend on how
the database backend is configured, so you can't make default values in
slapd.conf, because it depends entirely on how the user has created their
OpenLDAP is not a simplistic piece of software that you can just "install
and run". It requires in-depth knowledge of OS's, DB's, and a little bit
Principal Software Developer
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html