[Cerowrt-devel] Packet Pacing
dave.taht at gmail.com
Mon Feb 24 11:55:17 EST 2014
On Mon, Feb 24, 2014 at 10:13 AM, Iain R. Learmonth <irl at fsfe.org> wrote:
> I am attempting to perform some measurements using CoDel, PIE and
> DropTail using the CeroWRT platform but the rate-limiting appears to be
> happening before the queuing. The problem here is that there's no
> bottleneck between the queue and the output and so the AQM isn't really
> doing anything.
Huh? The SQM system creates a bottleneck and then various selectable AQM
technologies manage the queue in front, which is exactly what you want.
Note: In terms of strict pacing, the htb rate limiter system uses an increasing
quantum size to release bigger bursts of packets as we go to higher speeds.
(I think it starts increasing the quantum > MTU at about 20Mbit) -
this is because
the cpu scheduler can't effectively deliver single packets and still be useful
for other stuff.
see /usr/lib/sqm/ for details... or the gui...
If instead you want to run at line rate (10 or 100Mbit) there is
ethtool for that.
> I'd like to pace packets being dequeued to allow for the creation of a
> bottleneck where the output is the same link speed as the upstream link.
You can do this in multiple ways.
> Any ideas?
> urn:x-human:Iain R. Learmonth
> mailto:irl at fsfe.org
> xmpp:irl at jabber.fsfe.org
> GPG Fingerprint: 1F72 607C 5FF2 CCD5 3F01 600D 56FF 9EA4 E984 6C49
> Please verify out-of-band before trusting with sensitive information.
> [[[ To any GCHQ or other security service agents reading my email: ]]]
> [[[ Please consider if any professional body code of conduct to ]]]
> [[[ which you subscribe requires you to follow Snowden's example. ]]]
> [[[ Your professional membership, chartered or incorporated status ]]]
> [[[ may be at risk. ]]]
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html
More information about the Cerowrt-devel