Re: Severe netbsd-6 NFS server-side performance issues

[Tom Ivar Helbekkmo <tih%hamartun.priv.no@localhost>]

Martin Husemann <martin%duskware.de@localhost> writes:

> That reminds me of a strange problem I have on one machine - could you
> check vmstat -i output for the cpu timer interrupts?
>
> On my machine they usually are at 100 Hz, but sometimes after reboot
> I only get, say, 30 Hz - and that makes the network "not work too well".

Might this be the old timecounter TSC problem?

athene$ sysctl -a | grep timecounter
kern.timecounter.choice = TSC(q=3000, f=2400091650 Hz)
clockinterrupt(q=0, f=100 Hz) ichlpcib0(q=1000, f=3579545 Hz)
ACPI-Fast(q=1000, f=3579545 Hz) lapic(q=-100, f=266785816 Hz)
i8254(q=100, f=1193182 Hz) dummy(q=-1000000, f=1000000 Hz)
kern.timecounter.hardware = ichlpcib0

I've set it to ichlpcib0 in /etc/sysctl.conf, because it'll otherwise be
set to TSC on four out of five reboots, and the lucky fifth is because
the logic to select happens to glom onto the fact that TSC is quite
unreliable, and might start running slow any time.  (I could have set it
to "ACPI-Fast", but that seems to be called "ACPI-Safe" now and then, so
it's not a name to rely on.)

I've had to do the equivalent of this on all my machines.

-tih
-- 
"The market" is a bunch of 28-year-olds who don't know anything. --Paul Krugman