General Multiplexed ECN Channel

overview - Project description
papers - Documents
people - Who are we?


This project explores the potential of assigning multiple functionalities with some fields in a protocol header to allow adding new applications while maintaining header compatibility.

For example, the stream of ECN bits of a flow creates a communication channel whose purpose is to allow routers to signal congestion to the end user. But, because the ECN bit is infrequently set (i.e. ECN bit streams have lots of zeros and few ones), the rate of information transfer over this channel is significantly smaller than its capacity.

This opens up the possibility of creating a ``Multiplexed ECN" (M-ECN) channel which can be used to signal a variety of information from network to users (or users to network), without interfering with the basic functionality of the ECN field. We design robust coding schemes which improve the utilization of the M-ECN channel by exploiting the following key features of the ECN field: (i) ECN bits can be exchanged between two packets that are closely spaced in time at the expense of only a slightly delayed reaction at the sender, (ii) ECN bits are infrequently set; therefore, the availability of an unset bit close to a set bit makes it possible to swap them and construct code-words useful for signaling, (iii) a cluster of set bits is semantically equivalent to a single set bit because TCP reacts only to one decrease signal in a congestion window.

We apply the M-ECN channel abstraction to address an important problem in wireless networks: how to explicitly signal random packet losses to TCP sources so that they can distinguish them from packet drops due to congestion. Previous work has demonstrated that such signaling can lead to significant improvements in the utilization of wireless links.

We develop a mechanism, called WiSE, that uses a robust code to signal random packet losses over the M-ECN channel. Simulations show that it outperforms previous solutions while not harming non-WiSE sources or interfering with the ECN congestion signal. It achieves this performance boost without needing special fields in packet headers, or generating extra traffic.




Graduate Students:

NMS HomeProjectsPeoplePapersSoftware


M. I. T. Computer Science and Artificial Intelligence Laboratory · 32 Vassar Street · Cambridge, MA 02139 · USA