Date: Thu, 21 Apr 2005 19:30:11 -0700 From: "David Schwartz" <davids@webmaster.com> To: <freebsd-stable@freebsd.org> Subject: RE: Tuning for router performance Message-ID: <MDEHLPKNGKAHNMBLJOLKAEAMDFAB.davids@webmaster.com> In-Reply-To: <200504201407.j3KE7xNb000651@tuscan.xciv.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> In article <MDEHLPKNGKAHNMBLJOLKEEKADDAB.davids@webmaster.com>,
> davids@webmaster.com ("David Schwartz") writes:
>
> >> options HZ=1000 #for polling
> > That's too low. 2000 is the minimum you should consider.
>
> Having fixed my traffic generator, I'm now hitting ~220Kpps as limit
> before errors.
Excellent.
> > kern.random.sys.harvest.ethernet=0
> > kern.random.sys.hervest.interrupt=0
>
> Doesn't seem to make any real difference in my test.
That's strange, maybe it's because in polling mode there aren't many
interrupts.
> > kern.ipc.nmbclusters="262144"
> > vm.kmem_size="536870912"
> > vm.kmem_size_max="536870912"
>
> These don't really make any difference, I'm guessing would only help
> larger packet sizes?
The 'nmbclusters' change really only affects things if you have a lot of
local endpoints. And the kmem_size stuff only affects the case where you
have a lot of dummynet pipes or dynamic firewall rules. They're just part of
my typical router tuning because people often want these things as well.
DS
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?MDEHLPKNGKAHNMBLJOLKAEAMDFAB.davids>