These are the old pages from the weblog as they were published at Cornell. Visit for up-to-date entries.

June 18, 2003

Day 1 at Middleware 2003

It is hard to attend a conference while looking out over Copacabana beach.

Session 1 on Peer-to-Peer Middleware.

Ben Zhao presented a framework for using DHT based systems to maintain signature sets of Spam messages. The idea is that mail systems/reader identifies a spam message and then generates a signature set for that message, and stores each of the signatures in the DHT. Other uses can check signature sets of incoming messages to see whether a certain subset of the signature set matches with information in the network, which indicates a message can be identified as spam. Ben presented some work on reducing bandwidth and latency of the queries on the DHT by limiting the number of hops you can traveling on the network. He presented the impact of limiting the number of steps on the success rate of finding the correct answer. There are many questions about how to make this system work in practice. Ranging from implementing this at client vs the server, how to avoid erroneous or malicious signature insertions into the network, etc.

Second paper in this session was on the Narada-Broker from Indiana University. The presenter was not one of the researchers so the more interesting questions remained unanswered. The promote unlimited scalability of JMS style messaging, by grouping brokers (message servers). Many unanswered questions ...

Session 2 on Pub/Sub Middleware

The first paper is on event composition for pub/sub systems by Peter Pietzuch of Jean Bacon's Cambridge University group. The problem they are trying to solve is to have a pub/sub system generate new events based on pattern of events that have been generated. For example in the case of an active badge system, the composite even could be 'notify me when both Alice and Bob are in the kitchen'. The paper is mainly about how to define a language for composing event and how to implement it. The language has construct for recognizing time based event relations, recurring events, alternating event, etc.

The second paper is mobility support for pub/sub network by Ludger Fiege from TU Darmstadt. He looked at what the pattern is when a mobile user moves to a different broker. The basic pattern is to make a new subscription, get a state transfer from the previous broker and unsubscribe from there. The research looks at what support you can put in the pub/sub engines to handle mobility at system level instead of application level. Ludger demonstrated a few interesting mechanisms for predicting the user's mobility and proactively moving sub

Missing presenters.

Surprisingly there are two presenters of papers (Duke & Princeton) that didn't show up. In my book it means that your chairman or manager gets an angry letter from the organization. It certainly cannot happen for any prestigious conference.

Posted by Werner Vogels at June 18, 2003 02:12 PM