Virtual network mapping method based on deep reinforcement learning

Abstract

The invention discloses a virtual network mapping method based on deep reinforcement learning. The virtual network mapping method comprises the following steps: 1, modeling a node mapping problem in virtual network mapping into a Markov decision process; 2, on the basis of a Markov decision process, mapping virtual nodes by utilizing a DDQN algorithm; 3, performing virtual network link mapping byusing a shortest path algorithm; and 4, updating physical network resources, including CPU resources and link bandwidth resources. Through the self-adaptive learning scheme, a globally optimal mappingmethod can be obtained by saving energy consumption and improving the VNR receiving rate, and compared with a traditional method, the virtual network mapping method has better flexibility. Experiments show that the virtual network mapping method can reduce the energy consumption, improve the request acceptance rate and improve the long-term average income.

Description

technical field [0001] The invention relates to a virtual network mapping problem in a software-defined network, in particular to a virtual network mapping method based on deep reinforcement learning. Background technique [0002] With the rapid development of cloud computing, mobile Internet and other technologies, the demand for multi-tenant networks is increasing and flexible. The traditional IP-based framework has problems such as poor scalability and single core functions, which cannot meet the requirements of multi-tenant networks. Network business requirements. [0003] Network virtualization technology is an effective way to solve the above problems. It can integrate the existing virtualization technology of communication and computing resources, and adopt layered technical means to solve this problem. It is one of the key features that the future Internet should have. Network virtualization technology uses decoupling and multiplexing methods to share the physical r...

AI Technical Summary

Problems solved by technology

[0006] Since the traditional deep reinforcement learning cannot solve the inherent shortcoming of the Q-learning algorithm - overestimation, overestimation means that the estimated value function is larger than the real value function. If the overestimation is uniform in all states, then According to the greedy strategy, the maximum action of the value function can still be found, but the overestimation is often not uniform in each state, so the overestimation will affect the policy decision, resulting in the acquisition of an optimal strategy

Images