On Mon, 27 Jan 2003 jordan.breeding@attbi.com wrote:
> > On Mon, 27 Jan 2003 16:34:22 -0500 (EST)
> > Bill Davidsen <davidsen@tmr.com> wrote:
> >
> > > That's the thing I would expect to see if you used 'noapic' and watchdog.
> > > I posted over the weekend that I have been seeing some inobvious results
> > > to IPC benchmarking with scombinations of noapic and watchdog, but I
> > > didn't snap the interrupts. I'll be happy to add that to my list of stuff
> > > to try next weekend, but the box is not a toy during the week, and I would
> > > have a five hour round trip drive if a reboot failed, so I'll pass on
> > > trying it until I'll in the same room.
> >
> > no "noapic" here:
> >
> > kernel /boot/linux-2.5.59-mm5 root=/dev/hda5 ro vga=0x30a profile=2
> > nmi_watchdog=1
> There was a seperate thread about this a couple of days ago. It boils
> down to this: if there isn't sufficient load on your CPUs then it is
> both easier and faster to have one CPU handle all of the interrupts, if
> there is sufficient load on your CPUs then the kirq patch distributes
> the interrupts accordingly. I can verify that this behavior is
> consisent with what I see on my dual Xeon box (shows up as 4 ht capable
> CPUs). As long as your second CPU is showing up in /proc/cpuinfo and in
> top then the CPU is working it just isn't having to handle any
> interrupts yet.
The reason I mentioned noapic is that when running with the apic I see a
very good balance on a system with minimal load other than CPU. Since
there was a count on NMI the watchdog was clearly on. But on my test
system, I just check the count and the balance is very good:
Script started on Mon Jan 27 22:38:37 2003
bash-2.05a$ uname -a
Linux bilbo.tmr.com 2.5.59 #6 SMP Sat Jan 25 19:46:18 EST 2003 i686 unknown
bash-2.05a$ cat /proc/interrupts
CPU0 CPU1
0: 88705542 91864773 IO-APIC-edge timer
1: 853 850 IO-APIC-edge i8042
2: 0 0 XT-PIC cascade
3: 7752 7269 IO-APIC-edge NE2000
8: 1 1 IO-APIC-edge rtc
9: 0 0 IO-APIC-level uhci-hcd
11: 24218 28443 IO-APIC-level ide2
12: 26 26 IO-APIC-edge i8042
14: 664226 665346 IO-APIC-edge ide0
15: 1 1 IO-APIC-edge ide1
NMI: 0 0
LOC: 180596403 180596410
ERR: 25
MIS: 0
bash-2.05a$ exit
Script done on Mon Jan 27 22:39:00 2003
The system is up >24hr, so the irq counts reflect very little io. Running
noapic obviously puts all the ints on CPU0.
-- bill davidsen <davidsen@tmr.com> CTO, TMR Associates, Inc Doing interesting things with little computers since 1979.- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Jan 31 2003 - 22:00:18 EST