EXPLORING THE ALTERNATIVE WAYS OF INTERNET ROUTING:

IP MULTICAST

Alexandra Fedorova

Internet routing model specifies how the data is sent, or routed, over the Internet when one machine requests the data from another. The presently used routing model, known as unicast, handles one request at a time. If, for example, one million machines request exactly the same data simultaneously from the server, the server has to send exactly one million copies of the same data keeping itself extremely busy and guaranteeing a slow response to the clients. Multicast routing model allows increasing the efficiency of routing as it enables the server to take advantage of the situation when multiple clients request the same data. A busy server that is capable of sending data via the multicast channels would not send the data to every requesting machine. Rather it would send out the data only once and then the burden on copying the data would be distributed among the mrouters, the Internet routers that support multicasting. Steve Deering of Stanford first described the idea of multicast in his Ph.D. dissertation.

The goal of the project that I carried out this summer in collaboration with my advisor Lixin Gao, Ph.D. was an attempt to create a working application that would take advantage of the new routing model and uncover its strengths and weaknesses. The application was implemented in a new object oriented programming language Java that allows for a certain degree of platform independence.

The results were the working application Multicast Conference 1.0 (for detailed description and downloads go to http://cs.smith.edu/~afedorov/multicast.html). The Internet conferencing that is done with Multicast Conference 1.0 is carried out via the multicast channels; and the users can communicate by sending each other text messages and audio files.

The greatest weakness of the multicast routing model turned out to be the fact that it supports only an unreliable communication protocol, the UDP. The efficiency of transmission is compensated by the unreliability, so one probably would not want to use mbone for something other than multimedia. Research that aims to design reliable protocols for multicast is under way and so is the work to improve the current state of the mbone. (Supported by Schultz Science Foundation)

Advisor: Lixin Gao