The performance of Internet applications such as Web and Video depends critically on the underlying network conditions. Despite several advances in networking technology, the performance problems of Internet applications have not been fully alleviated. This is because, given the variability in the network and the application workload, no one protocol or parameter setting works best at all times. The goal of this project is to develop a novel, unified, and practical modeling framework that can accurately predict the performance of applications under different network conditions and parameters.
The project will achieve its stated goals by first developing a new congestion-aware TCP (Transmission Control Protocol) performance model, and then extending this model to encompass the application layer. At the TCP layer, performance will be modeled analytically, but will be augmented by empirically observed relationship between congestion in the network and the delivery rate. The model will use predictive and clustering techniques bootstrap model parameters based on network information from existing or historical flows in the vicinity of the target flow. For Web, this TCP model will be extended to predict page load times while for video applications, TCP throughput predictions will be used to drive the bitrate selection algorithm.
This research targets a broad segment of the population since video and web traffic makes up almost all of the traffic on the Internet. This work is beneficial to startups and smaller companies by providing means to systematically understand the effect of different protocols to improve performance. The results of this project will be used to improve the Web performance in developing regions as part of an ongoing collaboration. The PIs will introduce projects based on Web and Video performance, network protocol modeling, and measurement studies to undergraduate and high school students. The PIs have a successful track record of engaging high-school students and undergraduates in their research.
The software designed as part of this project will be released on a public site www.cs.stonybrook.edu/~arunab/econ. Electronic data will be archived internally. All data will be kept for at least 5 years after publication; likely longer, depending on its size and outside interest.
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.
