Network congestion avoidance through probabilistic packet injection scheduling based on mixed minimax strategy

Abstract

Tree-saturation in lossless network significantly degrades network throughput. In this paper, we propose Probabilistic Packet Injection Scheduling Protocol (PPISP), a congestion control mechanism in large-scale systems based on probabilistic injection to avoid hot-spot traffic. PPISP incorporates mixed minimax strategy, in game theory, to dynamically adjust packet injection probability. Using queuing delay at network buffer, PPISP detect congestion created at network and estimates the number of sources competing at congested channel to verify how many players exists in the game. Our simulation result showed that PPISP achieves 31\% improvement in throughput on hot-spot traffic pattern. Moreover, PPISP stabilizes network performance of background Uniform Random traffic on multiple hot-spot traffic pattern.