[nSLUG] FreeBSD 4.11 vs Redhat 5
D G Teed
donald.teed at gmail.com
Thu Mar 6 22:01:29 AST 2008
Two MX mail servers at my work happened to get 800 incoming
messages at about the same time. I graph the total queue and
active queue using cacti and spotted this excellence performance
Server A: Twin 3.2 Ghz Xeon with Redhat EL 5, postfix, amavisd and clamav
Server B: Twin 1.4 Ghz Xeon PIII with FreeBSD 4.11, postfix, amavisd and clamav
Server A took 45 minutes to process the active queue.
Server B took 20 minutes to process the active queue.
Both set ups have a very close match of versions and configurations.
The differences of software versions alone could only account for
something like a 20% swing. Actually, Server A has one software
config difference which should help it - the setting in amavisd
of "$bypass_decode_parts = 1" is set since clamd is said to be
better at handling that and it is mainly virii in the attachment we
are concerned with.
My thinking is that the fatter binaries of Redhat (SELinux, etc) are
hurting performance there. There might also be other OS factors like
kernel, etc. I mentioned the example of SELinux, not because it
is on, but because that is part of the server + kernel bloat present
in Redhat and not in FreeBSD. Even when not in use, the libraries
are still linked, the binaries are fatter and use more ram, etc.
I've heard people say BSD was leaner and meaner, and this result
plus my load averages seem to back that up.
Are there others who would have ideas as to what could
be differences I could do something about?
More information about the nSLUG