On Sun, Oct 05, 2008 at 01:02:20PM +0200, Rafael J. Wysocki (rjw@sisk.pl) wrote:
quoted text
> > Just for the reference, this one regresses from the 2.6.23.
> 
> Still, tbench results from 2.6.26 are better than from 2.6.27-rc.
> That's why it's still on the list, but if you think that's not relevant, please
> let me know and I'll drop it.

I meant it continuously regresses from .23 including 26-27 timeframe.
There were at least two changes in 26-27 which caused tbench regression
in this particular time interval. One of them enabled TSO over loopback
by default, so one can turn it off via ethtools or this patch:
http://marc.info/?l=linux-netdev&m=122263663631600&w=2

another (hopefully) one I'm trying to find out, which is a bit complex
and slow, since e1000 driver changes in the middle of the 26-27
timeframe does not allow my testing machines to boot.

So, there is a regression (partially resolved), but it is only part of
the problem.

-- 
	Evgeniy Polyakov
--
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