Nalanda

Citation: Wetherall, D. 1999. Active network vision and reality: lessons from a capsule-based system. In Proceedings of the Seventeenth ACM Symposium on Operating Systems Principles (Charleston, South Carolina, United States, December 12 - 15, 1999). SOSP ‘99. ACM, New York, NY, 64-79. DOI=http://doi.acm.org/10.1145/319151.319156

This paper presents an interesting idea: rather than have routing algorithms statically built into routers, allow applications to bundle their own algorithms along with their data packets, and have routers execute this application-determined code to make routing decisions. The combination of code and data is termed a capsule. Evaluation of the system is presented based on the ANTS toolkit.

Capsules declare a type in their headers, which is used to determine the code to be executed at the router. If this code is not already in the router’s cache, it is obtained from the neighbour which dispatched the capsule. This is an interesting choice of code distribution mechanism; as long as trust is not violated amongst the active routers, more explicit security checks on the code are needed only at the edge of the network. Indeed, security is a recurring theme in the paper, as it is a primary goal that untrusted users be capable of deploying code to the network without compromising the capacity of individual nodes, or of the network as a whole.

The idea presented here is certainly compelling as a thought experiment, but I found it difficult to place in context without examples of real usage. The performance numbers presented, for instance, were confusing, as they relate to no real application; I would imagine that performance would vary wildly depending on the capsule code.