Date: Wed, 17 Mar 2004 06:07:05 +1100 From: Peter Jeremy <PeterJeremy@optushome.com.au> To: freebsd-hackers@freebsd.org Subject: Re: kernel activity Message-ID: <20040316190705.GM56509@cirb503493.alcatel.com.au> In-Reply-To: <20040316084225.GA55231@icomag.de> References: <20040315150438.GA48241@icomag.de> <20040316070601.GK56509@cirb503493.alcatel.com.au> <20040316084225.GA55231@icomag.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Please don't top-post. On Tue, Mar 16, 2004 at 09:42:25AM +0100, Bogdan TARU wrote: > Thanks for the mails & advices... The box is a dual xeon @3GHz, with > 4GB of ram and raid 5 on board (scsi HDDs), with a 4.9 on it. Not short of horsepower then. 250K syscalls/sec may not be overly excessive for such a beast. Note that the 4.9 kernel is not re-entrant: only one CPU can be in the kernel at once. This will magnify the impact of high kernel load. > The box has two > NICs, one of them is a fxp with link0 activated (cannot use polling > because I don't want to give up SMP -- the userland activity is > already 40%, so giving up one CPU as to reduce sys load it's just > gonna leave the bottleneck where it is -- CPU, that is), You might be able to use polling with SMP - some people say it works but I think Luigi has some concerns - try rummaging through the archives. > other one is an em, but cannot use it since i don't have a gb > switch. Some em chips can run at 100Mbps - I presume you don't have a suitable interconnect. > Before activating link0 on fxp, the level of interrups/sec on > this interface peaked 6k, but after activating link0 it was reduced > to 2k. Still, a lot of sys activity... Have a look at "systat -v 1" or "top" and see if the system time is really interrupt time. If so, your only real option is a more efficient NIC and/or polling. > As a webserver, I run apache, stripped down from the modules that I > don't need, and compiled in php and some other modules > (statically). Most of the content that I serve is static, there are > only a few php scripts and they don't get much hits. I believe Apache2 can run in a multi-threaded mode rather than the multi-process mode that Apache1 uses. I presume you're not using multi-threading - if you are, you might like to compile it out and see what happens (threading adds quite a number of additional syscalls) > And no, I haven't tried turning HTT on and off, should I do that? The 2nd (virtual) CPU only provides a subset of the main CPU functionality and doesn't help kernel performance at all. It's possible that you will get better performance with it disabled. All I can suggest is trying it to see. > I am also considering trussing one of the apaches, to see what system > calls it's doing... I prefer ktrace(1) but this may be a reasonable idea. If all else fails, a second box with load balancing has the added bonus of providing you with some redundancy if either box fails. Your other possible option is 5.2.1 - 5.x can make much better use of multiple CPUs. If you go down this path, test it well and make sure that 5.x works for you. Peter
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040316190705.GM56509>