Performance Evaluation of Modified QUIC Protocol for Congestion Control Scheme
Deep learning based congestion prediction reduces network overhead to almost 10% level compared to handshaking based communication requirement for sending congestion state of the network amongst communicating nodes. The rerouting occurs to avoid congestion only when congestion state is detected. The congestion directly relates to dropping of packets and based on number of acknowledgements lost congestion state is finalized. Also, to reduce the congestion each node connected through common path reduces the communication data rate based on window management technique. In transmission control protocol (TCP) the slow start is commonly faced problem and also, due to reduction in window size due to congestion state, this slow start state occurs multiple times and overall data communication speed reduces. On the other hand due to less number of acknowledgements in QUIC protocol by google, congestion state information propagation requires additional control packets which increase the overhead. In this paper, the work consists of congestion state prediction using deep learning architecture and avoiding the congestion state with adaptive window management technique. The overall results analysis shows the improvement in communication speed over basic QUIC and TCP protocols along with almost only 10% of the network overhead.