CNS Core: Medium: Design and Deployment of Bespoke Congestion Control Algorithms

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Internet services are critical for US industry and government, the economy, health, education and social interaction. However, Internet services are not always timely or efficient, because Internet networks sometimes get congested. When Internet services transmit data, they are faced with a challenging question: how fast should they send? If they send too fast, the network will drop their data, leading to poor application performance like lost video frames or slow webpage load times. But, if they send too slowly, their data will arrive at the user slowly leading to poor application performance anyway. To make this challenging decision, applications use algorithms called Congestion Control Algorithms (CCAs) to determine the right sending rate: not too fast, and not too slow. This project is exploring new CCAs that are customized to applications; for example, applications involving video will be different than those that do not. If successful, applications performance over the Internet will be greatly enhanced. However, when several different CCAs are deployed over the Internet, they may compete with each other, and performance of some or all applications may suffer. The overarching goal is to enable application developers develop new, customized CCAs that provide better application experiences while still sharing the network equitably when different services are running at the same time. To this end, the project will design a new ultra-low latency bCCA tailored to online video game services using cross-layer optimizations that integrate bitrate adaptation with congestion control. It will perform large-scale measurement studies of the deployment of existing bCCAs, using novel machine learning approaches to identify those bCCAs that are already in use on the Internet. The project will empirically test whether new bCCAs and existing bCCAs share Internet bandwidth fairly. Finally, it will provide recommendations and guidelines for the design of future CCAs that can achieve application-performance objectives while also avoiding unfriendly behavior to competing Internet connections. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.