On Sat, 2009-05-23 at 16:35 +0200, Michael Riepe wrote:
quoted text
> Hi!
> 
> Michael Buesch wrote:
> 
> > Thanks a lot, Dave! This fixes the issue on my chip.
> 
> Yep, it's stable here as well. And even a little faster than pci=nomsi.
> The only strangeness I observed is that the throughput (measured with
> iperf and a single TCP connection) varies:
> 
> [ ID] Interval       Transfer     Bandwidth
> [  3]  0.0-10.0 sec    667 MBytes    559 Mbits/sec
> [  3] 10.0-20.0 sec    803 MBytes    673 Mbits/sec
> [  3] 20.0-30.0 sec    802 MBytes    673 Mbits/sec
> [  3] 30.0-40.0 sec    714 MBytes    599 Mbits/sec
> [  3] 40.0-50.0 sec    669 MBytes    561 Mbits/sec
> [  3] 50.0-60.0 sec    791 MBytes    663 Mbits/sec
> [  3]  0.0-60.0 sec  4.34 GBytes    622 Mbits/sec
> 
[snip]
quoted text
> I suppose it's a side effect of the MSI acknowledgement loop. But who am
> I to complain about higher average throughput? ;-)

I wonder if that is the TCP sawtooth pattern -- run up until we drop
packets, drop off, repeat. I thought newer congestion algorithms would
help with that, but I've not kept up, this may be another red-herring --
like the bisection into genirq.

A tcpdump may answer the question -- wireshark can do an analysis and
see if it is running up until it starts dropping or something.

Or it may be the loop, but I wouldn't expect it to make such a big
difference, or be as variable if it does.

Also, what does it look like with multiple streams?

Thanks for testing guys -- I'm glad it works for a sample size > 1!

Dave

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html