Novel congestion control method combining deep reinforcement learning and traditional congestion control

Abstract

The invention discloses a novel congestion control method combining deep reinforcement learning and traditional congestion control, and the method comprises the following steps: 1, enabling a transmission layer protocol to obtain performance preferences defined by practical application according to demands, and delivering the performance preferences to a deep reinforcement learning module of a novel congestion control algorithm; 2, collecting and updating round-trip time delay and packet loss probability according to response feedback by taking a predefined interval as a basic unit to reflectinformation of an actual condition in the network; 3, based on the existing information, respectively operating a deep reinforcement learning module and a traditional congestion control module, and obtaining congestion rate adjustment amounts decided by the two modules; and 4, selecting and calculating a final congestion window adjustment decision according to a predefined combination strategy, adjusting the current transmission rate, and transmitting data. The method can effectively adapt to application requirements, improves the performance, can provide better fairness and convergence, and has a practical value.

Description

technical field [0001] The invention belongs to the technical field of communication network congestion control, and in particular relates to a novel congestion control method combining deep reinforcement learning and traditional congestion control. Background technique [0002] With the continuous growth of WAN link bandwidth, the emergence of 4G network and 5G network applications and data centers and other scenarios, the industry's demand for congestion control algorithms has existed for a long time since it was proposed in the 1980s. Today's various network applications often have different requirements, such as low latency required for web browsing, high bandwidth required for downloading or video transmission applications, etc. A single congestion control algorithm is often difficult to adapt to all scenarios and application requirements, so that today's Linux There are more than fifteen congestion control algorithms built into the system kernel to choose from. [000...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the existing communication network transmission layer congestion control strategy cannot take into account the requirements of high performance, adaptability, fairness and convergence, etc., the present invention provides a new congestion control method that combines deep reinforcement learning and traditional congestion control. Reinforcement learning and traditional congestion control work together to make decisions together, absorb the advantages of both, solve existing problems, effectively adapt to the needs of applications, and finally improve performance; and in the actual situation where CUBIC is widely deployed in the current network, it can provide more Excellent fairness and convergence, with practical value

Images