Re: bdflush/rpciod high CPU utilization, profile does not make sense

[Jakob Oestergaard <jakob@unthought.net>]

On Tue, Apr 12, 2005 at 11:03:29AM +1000, Greg Banks wrote:
> On Tue, 2005-04-12 at 01:42, Jakob Oestergaard wrote:
> > Yes, as far as I know - the Broadcom Tigeon3 driver does not have the
> > option of enabling/disabling RX polling (if we agree that is what we're
> > talking about), but looking in tg3.c it seems that it *always*
> > unconditionally uses NAPI...
> 
> I've whined and moaned about this in the past, but for all its
> faults NAPI on tg3 doesn't lose packets.  It does cause a huge
> increase in irq cpu time on multiple fast CPUs.  What irq rate
> are you seeing?

Around 20.000 interrupts per second during the large write, on the IRQ
where eth0 is (this is not shared with anything else).

[sparrow:joe] $ cat /proc/interrupts
           CPU0       CPU1
...
169:    3853488  412570512   IO-APIC-level  eth0
...


But still, guys, it is the *same* server with tg3 that runs well with a
2.4 client but poorly with a 2.6 client.

Maybe I'm just staring myself blind at this, but I can't see how a
general problem on the server (such as packet loss, latency or whatever)
would cause no problems with a 2.4 client but major problems with a 2.6
client.

-- 

 / jakob