Distributed resource management continues to remain a fundamental challenge to the operation, development, and success of the Internet. The control of resources, such as network capacity, at topologically distinct locations is necessary in a variety of scenarios: managing complex peering agreements between ISPs, meeting quality of service requirements, delivering content from multiple locations, or defending against malicious network activity. The PIs are developing a new form of traffic policing called distributed rate limiting (DRL) that enforce global rate limits across traffic aggregates at multiple sites, enabling the control of distributed resources as if they were centralized. Critically, DRL ensures that transport-layer flows behave as if they traversed a single rate limiter in the Internet; DRL is effectively transparent, emulating the behavior of a central limiter.
The rate and amount of communication between limiters directly impacts the effectiveness of the system; DRL's accuracy and responsiveness depends on the underlying communication fabric. This project studies the ability of various communication fabrics (e.g., mesh, trees, and gossip) to deliver the requisite accuracy and scalability for distributed rate limiting. While DRL has immediate applications in distributed traffic policing, network admission control, and managing complex ISP traffic matrices, the research additionally explores the possibility of both enforcing different kinds of centralized behavior and controlling other rate-based resources.
Broader Impacts: The outcomes of this proposal include a public release of the distributed rate limiting mechanism for research, academic and non-commercial purposes. The PIs expect the resulting software to be especially useful in supporting undergraduate course projects.
