Optimization of the Performance Gap in Internet TCP/IP Routing
DescriptionThis project studies the performance difference in Internet TCP/IP protocol when either single-path routes or multipath routes are used. Our performance metric is the total utility achieved by jointly optimizing transmission rates using TCP congestion control and paths using IP source routing. In general, this performance difference is very hard to determine and approximate - in fact an NP-hard problem. Knowing when this performance gap is non-zero is important to the stability of TCP algorithms. We propose to use graph theory and optimization techniques to investigate when this gap is non-zero for a given topology and develop bounds to this "cost of not splitting". We propose another method motivated by recent breakthrough in sparse signal recovery to characterize the single-path solution and the performance gap. We also propose a computationally fast algorithm that can be combined with a branch-and-bound method to provide progressively tighter bounds to the performance loss.
|Effective start/end date||1/05/11 → 26/04/13|