From: T Scott Saponas (ssaponas@cs.washington.edu)
Date: Tue Oct 05 2004 - 22:45:37 PDT
"A Protocol for Packet Network Intercommunication" by Cerf & Kahn - Reviewed by T. Scott Saponas.
This paper presents a protocol for internetworking packet switched networks. It suggests mechanisms for handling addressing, routing, multiple simultaneous reliable connections over unreliable networks, variations in packet sizes, and flow control.
The paper lays out much of what would be needed to connect packet switched networks of different types (radio, wired, etc) and of different administrative domains. It does this well by addressing many different challenges posed to these networks. Particularly, it is suggested only a single well known and fairly unassuming protocol could be used across all networks such that gateways would not be burdened with converting between protocols. Similarly, it suggests dealing with different packet size requirements of different networks gateways might need to fragment packets but should not be burdened with the memory intensive task of de-fragmenting.
Some of the limitations of this paper are: how it suggests doing accounting is hand-wavy, there is no discussion of data security - only data integrity, it is stated 256 networks and 65,536 TCP's per network are enough (there is never enough of anything in CS), congestion control doesn't seem to be addressed.
The relevance of this paper today is the philosophy put forth for internetworking. The ideas like fragmentation and one protocol as the center of the hourglass seem like the enabling ideas for internetworking among different types of networks and different administrative domains. This paper leaves room for some future work in internetworking as far as some exact mechanisms for accounting, dealing with the tension between applications that want reliable vs best effort data transmission, congestion control, and security.
This archive was generated by hypermail 2.1.6 : Tue Oct 05 2004 - 22:45:38 PDT