SmartRE: An Architecture for Coordinated Network-wide Redundancy Elimination, Ashok Anand, Vyas Sekar, Aditya Akella, in Proc. of ACM SIGCOMM'09, August 2009, Barcelona, Spain.
Traffic in network exhibits some redundancy and this paper presents an architecture which attempts to encode packets in order to remove these redundancies. It take into account some specific constraints (encoding and decoding rates) and defines where to cache the packets and how to ensure that an encoded packet based on the information at a router will find a decoder further down the path.
The proposed approach approximates relatively well an omniscient encoder and removes most of the redundancy. It also demonstrates that handling network-wide redundancy removal is preferable to hop-by-hop.
While the approach is promising, it assumes that there is redundancy in the traffic. There are some WAN optimization products and thus a market, but some of the evaluation assumes a 10% redundancy. As such, setting up a complex caching and encoding/decoding infrastructure for such a gain does not seem worth it. I would like to see better characterized the true redundancy gain, before dwelling on how to remove these redundancies.
I attended Sigcomm in BCN, and I don't recall seeing this talk, even though I most likely have. It was in a morning session, and I don't think I missed any morning session. Funny how remembering the talk would have saved me from re-discovering these results.