[Date Prev][Date Next]
Re: Casual benchmarking OS performace with OpenLDAP
Certainly it seems something fishy is going on here, but it's hard to tell.
James F. Hranicky wrote:
The DB library would be my first suspect; check and see if it has a
dependency on -lpthread. We usually build BDB with fast gcc/asm mutexes
but they may have built it with posixmutex support instead, which is
significantly slower. In 4.2 the gcc mutexes are the default, but I have
no idea how Debian configures their build.
I'm in the process of trying to find the best OS platform on which to run
OpenLDAP as a backend for Heimdal and Samba, as well as a replacement for
Currently, I have a machine with a 2.8GHz Xeon and a 140G drive which
holds both Debian Linux (kernel 2.6.11) and Solaris 10 x86 .
Here is the pertinent server info:
OpenLDAP 2.3.2beta (gcc -O2)
DB 4.2.52 w patches (Debian default)
In general, you can't construct a test like this unless you know you're
doing apples-to-apples comparisons. If you built your own BDB library on
Solaris x86 but used the distro's stock BDB install on Linux, you really
have no idea if the two builds are comparable.
There are a lot of global directives interspersed with database-only
directives here. Generally we discourage this because it leads to false
assumptions about how things work. The slapd.conf(5) manpage (and
associated config manpages) are pretty explicit about what directives
are global vs not. In 2.3 with the LDIF configuration it will be firmly
delineated, since directives will only be valid in specific config entries.
OpenLDAP 2.3.2beta (SUNWspro/cc -xO4)
DB 4.2.52 w patches (SUNWspro/cc -xO4)
Here's the slapd.conf I'm using:
-- Howard Chu
Chief Architect, Symas Corp. Director, Highland Sun
Symas: Premier OpenSource Development and Support