Re: [patch] voluntary-preempt-2.6.8.1-P8

[Lee Revell]

On Mon, 2004-08-23 at 23:56, K.R. Foley wrote:
> Lee Revell wrote:
> > On Mon, 2004-08-23 at 17:01, Ingo Molnar wrote:
> > 
> > 
> >> - reduce netdev_max_backlog to 8 (Mark H Johnson)
> >>
> > 
> > 
> > On my system this setting has absolutely no effect on the skb related
> > latencies.  I tested setting netdev_max_backlog to every power of two 
> > between 1 and 128, and regardless of this setting, I can produce a
> > 450-600 usec latency with:
> > 
> > ping -s 65507 -f $DEFAULT_GATEWAY
> > 
> > Looks like skb_checksum is the problem.  Here is one of the traces:
> > 
> > http://krustophenia.net/testresults.php?dataset=2.6.8.1-P8#/var/www/2.6.8.1-P8/trace14.txt
> > 
> > Lee
> > 
> > -
> 
> Setting netdev_max_backlog doesn't seem to have any effect for me either.
> 
> I get a similar latency when I try to reproduce your result above, ~612 
> usec. However, my trace is very different than yours:
> 
> http://www.cybsft.com/testresults/2.6.8.1-P8/latency_trace3.txt
>

I have some that are similar to this.  Looks like many of the
differences are due to different network drivers (e100 vs via-rhine).

> In fact most any network activity, with the POSSIBLE exception of 
> already open connections, seem to be able to trigger higher latencies. 
> As you can see here:
> 
> http://www.cybsft.com/testresults/2.6.8.1-P8/
> 

Looks like except for the one above, all of these involve the
extract_entropy issue.  I am using a hack to disable the random driver
to avoid these latencies.

There are a few other threads going on discussing a real solution to the
extract_entropy issue.  However, none of the solutions that has been
posted yet fix the problem.

Lee


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/