This paper presents an end-to-end congestion control algorithm optimized to support background transfers. Surprisingly, an end-to-end protocol can nearly approximate the ideal router-prioritization strategy by (a) almost eliminating interference with demand flows and (b) reaping significant fractions of available spare network bandwidth.
Our Internet experiments suggest that there is a significant amount of spare capacity on a wide variety of Internet links. Nice provides a mechanism to improve application performance by harnessing this capacity in a non-interfering manner. Our case studies demonstrate that Nice can simplify application design by eliminating the need to hand-tune parameters to balance utilization and interference. Inspired by the results in this paper, we have built a self-tuning prefetching system [31] based on Nice that avoids interference at the server and in the network, and is deployable with simple modifications to a web server.
One application of Nice is to support massive replication of data and services, where spare resources (e.g., bandwidth, disk space, and processor cycles) are consumed to help humans be more productive. Massive replication systems should be designed as if bandwidth were essentially free. TCP Nice provides a reasonable approximation of such an abstraction.
=
currsize0.85