Distributed Dynamic Management System Based on Recurrent Neural Networks

Shu-tian ZHOU

Abstract


Distributed network system is responsible for many dynamic information computing services, and all tasks are scheduled through resource managers. Large-scale distributed network systems need to integrate heterogeneous and dynamic characteristics into the content of resource allocation and dynamic management. However, the diversity of network equipment and internal communication brings great difficulties to resource allocation/dynamic management. The effectiveness of scheduling decisions issued by resource managers is also difficult to determine. The resource allocation and dynamic management scheme of distributed network system proposed in this paper adopts bi-directional recurrent neural network learning to realize reliable and efficient resource allocation and operation and maintenance management solution. The output of bi-directional cyclic neural network is dependent on the current input and memory to synthetically decide when to match a task with a resource and improve the utilization and performance. Experiments in real distributed network systems show that the technical indicators of this method, including efficiency and execution rate, will not be significantly reduced by the increase in the number of tasks. Therefore, this method can further enhance the reliability and scalability of the resource allocation and dynamic management of distributed network systems.

Keywords


Distributed network system, Heterogeneous and dynamic characteristics, Bi-directional recurrent neural network


DOI
10.12783/dtssehs/emse2018/27234

Full Text:

PDF