As the accidental co-authors of what may well become the world’s largest ECN-enabled AQM deployment, our ecn-sane project is started in the hope that a clear, safe, sane path forward for ECN enablement across the edge of the Internet will emerge.
Explicit Congestion Notification is a means to do network congestion control without dropping packets.
Neither the original codel or pie AQM research covered ECN.
However, the fair queueing variants of these algorithms, the ‘fq_pie’ and ‘fq_codel’ qdiscs enabled ECN by default, because, in very limited tests by bufferbloat.net members in 2012, it seemed to work well. FQ_Codel, in particular, is in increasingly wide deployment. We’ve long encouraged individual users to try it out… and then, in 2017… Apple enabled it universally across their devices and stacks.
ECN’d behaviors are observably different than drop, and different AQMs treat it differently. The pie algorithm starts dropping ecn enabled packets once the drop probability cracks 10. RED behaves similarly. Codel does not drop until it runs out of packet space.
Our limited tests showed codel alone to be somewhat ineffective against ECN, and in both pie and codel’s single queue implementations in Linux we left it disabled by default. Cake developed a more refined approach to ECN management. The sqm-scripts enable ECN for inbound universally and disable it for outbound. Our FQ_Codel implementation for WiFi, now shipping in quantity millions, enables it universally. Inconsistencies in ECN behavior abound in both AQMs and TCPs.
Much of this project will be focused on analyzing and reducing any additional congestion caused by modern protocols with ecn enabled, as well as examining potential side-effects on other protocols.
We expect much work to take place on the mailing list. Like all bufferbloat.net lists, ecn-sane is an open mailing list, however, given the level of religious advocacy of ecn elsewhere on the Internet, it has several policies that are new to bufferbloat.net.
People will be banned, after 3 warnings, from the email list, for the following reasons:
Scientific skepticism of both negative and positive results is utterly required here.
We are hoping that these rules here are sufficient to keep the noise level low, and to make progress forward on this sensitive topic.