A Theory of Packet Flows Based on Law-of-Mass-Action Scheduling

Abstract

Designing dynamically robust protocols is not a simple task with current classic work-conserving scheduling, where packets are sent out as soon as processing and transmission capacity is available. We show that deviating from this fundamental queuing assumption leads to much more controllable and analyzable forms of protocols. At the core of our work is a queue-scheduling discipline based on the chemical "Law of Mass Action" (LoMA) that serves a queue with a rate proportional to its fill level. In this paper we introduce our LoMA-scheduling approach and provide a solid mathematical framework adopted from chemistry that simplifies the analysis of the corresponding queueing networks, including the prediction of the underlying protocols' dynamics. We demonstrate the elegance of our model by implementing and analyzing a TCP-compatible "chemical" congestion control algorithm C3A with only a few interacting queues (another novelty of our approach). We also show the application of our theory to gossip protocols, explain an effective implementation of the scheduler and discuss possibilities of how to integrate mass-action scheduling into traditional networking environments.