On Fri, 2007-10-05 at 18:59 +0200, Martijn Lievaart wrote: > Rik van Riel wrote: > > Somehow I suspect it will not work too well on my own list, which > > does around 400 queries/seconds/server, but having a faster way to > > get systems "on the list" is an excellent idea for smaller lists. > > > > It does about 700 queries/sec on my dual core Athlon. The bottleneck > seems to be CPU. Profiling revealed that a lot of time is spend > constructing and deconstructing the DNS packets. > > I'll go over the code in the weekend, to see if there are any obvious > > bugs, before I commit the thing. > > > > Wait sec before you commit it. I'm working on an optimised version. It > should be much faster. This version does 1000 query/sec running on a 1600Mhz Sempron. About 70% cpu for skbld and 30% mysqld. On a dual core Athlon (1800Mhz?) I get about 850 queries/sec where the benchmark code takes up about 150% of cpu and skbld about 50%. Maybe I'll benchmark it tomorrow on a 8 core Xeon. :-) So this version should be adequate for most installations. As it is still single threaded, adding CPUs will not make it run faster (except other code can run on other CPUs). Note that I benchmarked directly against the daemon, not going through named. The forwarding through named might make a difference, I'll test that as well soon. > > Speaking of which, I wonder if we should move the configuration to > > /etc/sysconfig/spamikaze (with /etc/spamikaze/config as fall-back) so > > we better fit in with most Linux distributions... > > > > > > Do more distros use this beside the RH derivatives? I thought about this, but /etc/sysconfig/ on RH contains information used by the init scripts. Which is also what I used in my redhat style init script. So I don't think this is a good idea. /etc/spamikaze is much better in my eyes. M4
Attachment:
skbld
Description: Perl program