SIGCOMM 2008

    SIGCOMM 2008 had about 600 people attend, and felt a little like one of the early AUUG conferences. One of the things I am learning as I get older is that I am having increasing trouble sitting in hard hotel meeting chairs all day -- my back just wont let me. Surprisingly, I find most university lecture theatre seats more comfortable.

    Below is a summary of the talks I thought are particularly good.

    Don Towsley keynote

    Don Towsley is the winner of the 2008 ACM SIGCOMM Award. His talk wasn't really a paper, he instead spoke about the need for models when performing Internet measurement research. If you don't have a theoretical model, then you don't have a way of verifying that your samples from the Internet are valid or not. This is an interesting point I hadn't considered, and which affects my own research work. My new problem is I'm not really sure how to produce models which have meaning to my current SMTP survey project. Additionally, Don asserts that PhD candidates shouldn't attempt to implement new applications. Instead they should work on enabling new applications. This observation is based on many years of supervising PhD candidates and their relative success.

    A Case for Adapting Channel Width in Wireless Networks

    Microsoft Research -- WiFi cards use a fixed bandwidth of 20 MHz. Using this quite large bandwidth for idle connections consumes a lot of power. This research proposes varying the size of the channel depending on what needs to be transferred -- 5MHz for an idle connection, up to 40 MHz for an active transfer. This does require a protocol for both sides of the connection to agree on a channel width. Sample implementation using the Zune peer to peer song sharing protocol.

    Spamming Botnets: Signatures and Characteristics

    Microsoft Research -- using URL extraction and a regular expression generator to find spam emails from botnets. Extract URLs from emails, track by domain over time and note bursty arrivals of such URLs. Send the bursty ones to a regular expression generator, which is then generalized to exclude victim specific IDs as well as domains, and then filter based on that. An interesting talk.

    To Filter or to Authorize: Network-Layer DoS Defense Against Multimillion-node Botnets

    University of California, Irvine -- DoS flooding attacks are a serious problem. The number of sources can be huge, as well as the packet count and bandwidth consumed. There are currently two schools of though on DoS protections -- filtering (anyone can send, and then filters are added when an attack occurs), and capability based systems (senders request permission before sending, and then use proof of that permission in each packet they send). This paper compares the two approaches.

    BitTorrent is an Auction: Analyzing and Improving BitTorrent’s Incentives

    University of Maryland -- the amount of research that is being conducted into peer to peer protocols, especially BitTorrent, is really interesting. This paper presented an alternative algorithm for how to select which blocks to offer for upload in return for the highest possible download rates. Specifically, it reframes BitTorrent as an auction system, in which leechers should be bidding the lowest possible in order to be selected as a download partner. This is implemented in BitTyrant (http://bittyrant.cs.washington.edu/). The rest of the talk focuses on strategies for gaming BitTorrent based on this observation. Very interesting. Implemented in a client at http://www.cs.umd.edu/projects/propshare/.

    Network Discovery from Passive Measurements

    UW Madison -- the traditional approach to mapping the Internet is to use active measurements. This paper proposes a solution using passive measurements. The underlying problem is that you have data in the form (source IP, destination IP, number of hops) and from that you need to determine which hops are in common for any given pair of readings.

    Taming the Torrent: A Practical Approach to Reducing Cross-ISP Traffic in Peer-to-Peer Systems

    Northwestern University -- this paper proposes piggy backing on CDN networks in order to determine which peer to peer clients are nearly to a p2p leecher. The network routing overlays produced by these networks can be used to select peers which can provide downloads more efficiently.

    Tags for this post: conference(S) sigcomm2008(S)

posted at: 15:48 | path: /conference/sigcomm2008 | permanent link to this entry