Re: [PATCH] net_sched: sfq: add optional RED on top of SFQ

On Fri, Jan 6, 2012 at 8:43 PM, Rick Jones [off-list ref] wrote:

On 01/06/2012 11:33 AM, Eric Dumazet wrote:

quoted

Le vendredi 06 janvier 2012 à 10:30 -0800, Rick Jones a écrit :

quoted

netperf nitpick :)  While I doubt that Dave Taht is running it that way,
one can have multiple requests in flight on a single _RR test via the
test-specific -b<additionaltrans>  option.  That option is enabled by
default (--enable-burst on the configure) in 2.5.0 and later.

Ah Rick, I dont think we can tune IP_TOS with netperf -t UDP_{STREAM|
RR} ?

I ask because it could be a good thing to set ECT(0) on datagrams to
check our ECN capabilities and get in the final report from receiver a
count/percentage of CE frames.

For other apps than netperf, being able to see this stuff in iptables
(ip6tables) might be helpful. Did this kernel patch series make it?

http://comments.gmane.org/gmane.comp.security.firewalls.netfilter.devel/41139

Funny you should mention that :)  In the top-of-trunk (perhaps it is in
2.5.0 too, I do not recall) there is the global -Y option:

$ src/netperf -Y
src/netperf: option requires an argument -- 'Y'

Usage: netperf [global options] -- [test options]

Global options:
...
   -y local,remote   Set the socket priority
   -Y local,remote   Set the IP_TOS. Use hexadecimal.

So long as you either use the omni code directly, or indirectly by not
undoing WANT_MIGRATION those should work - for some definition of work
anyway...I would not be surprised to learn there are bugs in the support.

Also in top of netperf trunk is a mode to be able to exercise different TCP
congestion control algorithms. I was mostly fiddling with westwood, tcp-lp and
tcp-ledbat v1 ( https://github.com/silviov/TCP-LEDBAT )

" The output selectors are LOCAL_CONG_CONTROL and
   REMOTE_CONG_CONTROL and setting is via the test-specific -K option."

I also have a rsync patch for priority, diffserv and congestion control
floating about...

I'm a little vague as to whether your related kernel patch for inheriting
the congestion algorithm made it in?

I note that netperf trunk breaks backward compatibility with
netperf 2.5.

However, there is nothing presently in the netperf code to cause any
*individual* send to be so marked independently of the others.

Heh.

happy benchmarking,

rick jones

-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
FR Tel: 0638645374
http://www.bufferbloat.net