Objective window adaptation for transport protocols under imperfect information

The transmission control protocol (TCP) window adaptation scheme is heuristic in nature. Previous work on such schemes has shown that window size calculations, based on utility-driven and game-theoretic objective functions, lead to better performance in overall throughput. We observe that packet losses are good indicators of network congestion. We utilize the trend in packet losses to propose a predictive utility-based scheme working under imperfect network information. In our work, we record the packet loss observed in the current and previous transmission slots to predict the expected packet loss for the next interval. We then use a simple utility-based scheme to calculate the best window size for the next interval that minimizes the packet loss and maximizes the net utility.