The Design And Implementation Of Resources Monitoring And Allocating System Based On Super Computing Integration Platform

As super computing technology developing and becoming mature,a large number of supercomputing cluster system appeared on the scene and become the mainstream platform for all the high performance computing technique.In these circumstances,real-time monitoring,dynamically allocating and rational utilizing resources are germane and central to the issues.For one thing,based on the telnet ways,traditional cluster monitoring system is easy realizing but inefficient,serious delay,poor visualization and can't use the monitor result effectively.For another thing,supercomputing cluster system adopt simply resource allocation policy su ch as first come first service and Max-min in resource management,these strategies is not only simple but also can't use the resources monitoring state data to guide resource allocation,which caused the disconnection between resource monitoring and allocation.These problems restricted the further improvement of user tasks execution,so we designed and implemented a system of resources monitoring and allocation in this paper.First of all,this paper proposes a resource allocation policy which compromis es resources monitoring on the base of modified ant colony algorithm IWO-ACO(Ant Colony Optimization based on Invasive Weed Optimization),in the view of resource allocation methods are too single,task execution efficiency is too low and did not make full use of resource monitoring system state data.Based on the “diffusion map” of invasive weed optimization algorithm,we improved the pheromone update rule of ant colony algorithm,which improved convergence rate and capacity of searching optimal solution;In addition,strategies such as dynamic evaporation,concentration control and dynamic equilibrium are adopted,which greatly avoid premature phenomenon.Finally,with the modified ant colony algorithm and load data of resources monitoring and allocation system,create a resource allocation strategy which fuse resources monitoring policy,corresponding process description and algorithm flowchart are given.Secondly,according to the IWO-ACO resource allocation strategy that based on integration of resource monitoring,this paper designed and implemented a resources monitoring and allocating system of supercomputing integration platform to make up for the drawbacks of monitoring delay and poor visualization.The system uses the QuartZ task scheduler of the Spring framework polling cluster load computing nodes,and then use image or curve form to display the real-time resource state of the supercomputing cluster to user,such as node state,CPU/GPU utilization and memory utilization etc.The resource monitoring system can provide resource status data as load data for the resource allocation policy in this paper.There is a circulating cooperative mode "positive feedback load-rational allocation of resources-resource monitoring" between the resource monitoring and resource allocation.Finally,this paper carried out some experiment and test.The results show that IWO-ACO resource allocation strategy can effectively improve the efficiency of task execution,reduce user task turnaround time and the average waiting time.Resource monitoring and allocating system of supercomputing integration platform provide friendly interface and real-time display of the cluster resources status.It has great application value.