Hierarchical distributed deployment method and system for virtual network based on reinforcement learning

Abstract

The invention discloses a hierarchical distributed deployment method and system for a virtual network based on reinforcement learning, which aim at network simulation based on a Docker containerized virtual network. Aiming at the problems of virtual network optimization, slicing and deployment, under a physical host machine distributed environment in which calculation, network and storage resources are limited, through a reinforcement learning framework, a reward mechanism considering physical host machine resource consumption and virtual network block cross-host machine communication performance loss is designed, and long-term benefits of different actions adopted by each physical host machine in different resource supply states are continuously calculated; an algorithm dynamically, autonomously and continuously learns an optimized virtual network slicing scale and a proper deployment opportunity, and certain dynamic randomness is introduced through a randomness strategy, so that the problem of over-consumption of resources or over-fragmentation of slicing caused by excessive and rigid selection of the maximum action is avoided; therefore, the purpose of hierarchical and distributed optimization deployment of the virtual network is achieved.

Description

Technical field [0001] The present invention relates to the field of network virtualization technology, specifically, to a method and system for a distributed deployment based on reinforcement learning virtual network hierarchy. Background technique [0002] Network simulation is a key support structure, protocols, algorithms, research computer network system. Since Docker containers retained substantially full TCP / IP protocol stack, with respect to the virtual machine, the container has a higher efficiency and less startup performance overhead, container technology became popular, but also provides a network simulation new idea - Docker containers to build the core virtual network element (e.g., a virtual router, a virtual switch, a virtual terminal system, etc.), with the use veth-pair, OVS (Open vSwitch), VxLAN (virtual eXtensible Local Area network) technology generating a virtual link connection, is formed to simulate virtual network deployed on the physical host. Since th...

AI Technical Summary

Problems solved by technology

The present invention is oriented to network simulation based on Docker containerized virtual network, aiming at the problem of virtual network optimization and deployment, in the distributed environment of physical host with limited computing, network and storage resources, through the reinforcement learning framework, the design takes into account the physical host Reward mechanism for host resource consumption and cross-host communication performance loss for virtual network slicing ,according to Continuously calculate the different actions taken by each physical host in different resource supply states long-term benefits , so that the algorithm according to Dynamically and autonomously continuously learn the optimized virtual network block size and appropriate deployment timing, and introduce a certain amount of dynamic randomness through the randomness strategy to avoid excessive rigid selection The biggest action will cause the problem of excessive resource consumption or fragmentation, so as to achieve the purpose of virtual network layering and distributed optimization deployment

Images