I have several mail relays on Sun Sparcs with Solaris 8 and 9 running exim, exiscan, and clamav. All the same version. top shows clamd uses 5%-40% CPU time (it was always among top 10 highest CPU users). I figured "hey, since this is a busy server then this load is normal". A rather interesting fact is that out 512 MB memory, top shows that more than half of it is unused,
Incidently, I stumbled accross Gentoo, and see that they have an up-to-date release for Sparc systems, complete with tftpboot image for Sparc64. I install it, upgrade the kernel to 2.6.9, upgrade glibc to nptl-enabled version, and install the necessary daemons. BTW, I have used linux on Sparc earlier on another server (Aurora Linux, RH 7.x clone, kernel 2.4) but that one was rather outdated.
I immediately notice something different : the system is now more responsive. A more interesting fact is that now clamd uses very little CPU time (top shows 0.0% often it is not in 35 highest CPU users, thus top don't display it). Another intersting change is that now almost all available memory is used, although over half of them is cache.
Any idea how this can happen? Is nptl really THAT much better then solaris threads? Or Linux is simply faster than Solaris? Or is it just because clamd runs so much better on linux than any other OS?
Right now I'm thinking about Gentoo-ing all Ultrasparc-II-based mail relays. Too bad Linux don't run on Ultrasparc III yet.
Regards,
Fajar _______________________________________________ http://lists.clamav.net/cgi-bin/mailman/listinfo/clamav-users
