I used to dislike LDAP, but I’ve since come to terms with it. The protocol itself is certainly not intuitive like some, but once you understand it, you realize that it can indeed make your life much easier in a lot of cases.

A good example of this would be using it as a database for accounts on servers (e.g. shell accounts, mail accounts, and the like). You can also do things like combine PAM with LDAP, and have applications authenticate users via PAM/system auth, which will in turn use LDAP —this means the application doesn’t really have to know about LDAP. While this is easy and good for simple setups, I have now decided that it is better to separate this out per-application, for finer access control.

LDAP also provides for replication, meaning you can have multiple servers with the same data (the slaves will be updated by the master). You can then use built-in functionality for trying a second server if the first fails, and you could also use something like CARP.

The only thing that can be annoying about LDAP is that some developers aren’t very clear about what exactly their program’s configuration directive does, so you have to experiment or go dig around in the code to figure out what it expects, does, and so on. This seems to be a bigger problem with LDAP than most other things. At least a lot of applications even support it though.

I did some tests to compare performance of a kernel resembling GENERIC against an SMP kernel with HyperThreading enabled. The results were not terribly surprising, but HT actually helped a lot more than I thought it would. Intel originally said that HyperThreading can increase performance in some situations by up to 20%, and I guess they were right. Of course this depends on the operating systems ability. FreeBSD 6.1 performed quite well in this case—I have heard HT actually hurts performance in some cases. For these tests, it did, but in a very minor way and only when 1 thread is used (no real surprise there).

I posted this to the freebsd-smp@ mailing list, and some of the people would like to see additional results, comparing performance when HT is disabled in the BIOS, switching timecounters, and using a couple of other benchmarking tools.

I will, I just have to find some time first…

The results are here.

Last night was maintenance night and I upgraded the servers, and put SMP kernels on them, since Alvis has been stable and everything else has too. It was quite a smooth process.

Today I am playing around with some statistics, specifically on our web proxies, using a tool called Calamaris. It parses log files and is able to gather a wide variety of statistics about proxy usage.

Facebook, WWU, and MySpace were at the top of the list on most-visited sites. No surprise there.

You can see some of the data here.