By: Mat Ford
Date: July 1, 2013
Back in the last millennium, the IETF published an informational document (RFC 2309) that provides two important recommendations concerning measures to improve and preserve Internet performance. The first of these recommendations concerned the need for active queue management (AQM) to reduce the number of dropped packets, and to provide better service for low-delay, interactive flows. The primary goal of AQM algorithms is to allow network operators simultaneously to achieve high throughput and low average delay by detecting incipient congestion. Random Early Detection (RED) was identified as an appropriate AQM algorithm to provide these benefits. The second recommendation in the document concerned the need to continue efforts to deal with flows that were unresponsive to congestion signals.
In the many years since this advice was published, the Internet has grown and diversified considerably, but the need to provide a way to balance the demands of throughput-maximising flows with those of latency-sensitive, real-time application flows has not diminished. Unfortunately, RED has not seen widespread adoption among the network operator community for a variety of reasons, including configuration complexity. As more and more real-time, interactive applications are deployed, the need keeps growing for better algorithms in routers to manage the queues that are a feature of packet networks.
The combination of work on real-time communications tools (e.g., WebRTC) that will bring real-time audio and video communication to browsers with the realization that many of the buffers in networking equipment are both overly large and completely unmanaged has resulted in renewed interest in providing recommendations to the Internet community on the topic of AQM. Almost simultaneously, new AQM algorithms have emerged that promise dramatically improved performance for interactive applications in the presence of competing, throughput-maximising flows. These algorithms trade a small amount of bandwidth for significant latency improvements, and offer the promise of little or no configuration for the operator.
During the Transport Services Area (tsvarea) open meeting in Orlando, the Area Directors led a discussion of whether it was time to charter new work on AQM in the IETF (http://www.ietf.org/proceedings/86/minutes/minutes-86-tsvarea). There was strong support for initiating work on the topic in the IETF, to work on documenting new AQM algorithms, developing informational documents and working with other areas to understand where similar issues arise. Work is already underway to revise RFC 2309 in light of the experience and developments of the intervening years since its initial publication, and a new mailing list for discussion of AQM developments has been formed (https://www.ietf.org/mailman/listinfo/aqm). A BoF proposal has been submitted for the IETF 87 meeting in Berlin where the discussion will continue and where a more detailed work plan will be developed.
Widespread deployment of configuration-free AQM holds the promise of an Internet where it is the exception, rather than the rule, that real-time and interactive applications like voice and video communications tools suffer in the presence of throughput-maximising file transfers and other bandwidth-intensive flows. This would be a significant step forward—IETF work in this area deserves our close attention.