Research On Load Balance Algorithms Based On PC Clusters

With the rapid development of grid and high quality-value ratio, PC clusters which are connected by high-speed LANs become the mainstream in scientific research and production. The key problem of clusters is how to manage efficiently and improve system resources utilization. Job management systems(JMS) has a good support for clusters scheduling and management. JMS provides many scheduling algorithms for different customers'needs.OpenPBS is a open-source product for famous JMS PBS, which is used for production and living widely. But the OpenPBS's default scheduling algorithm is FIFO, which is not effectively and can't prove the fairness of algorithm. In addition, the algorithm of OpenPBS belongs to the static load balance, which don't adapt to dynamic changes of resources in clusters scheduling. The methods which are simulating natural ecological mechanism for solving complex optimization problem become hot computational intelligence topics, especially Ant Colony Optimization algorithm (ACO) , which has become an important way of clusters scheduling. ACO has good dynamic currentness and information feedback ability, which is similar to the dynamic performance of the cluster resources and is proved to be an effective algorithm to solve scheduling problems.But the existing ant colony algorithm can not solve the scheduling problems liking misusing good performance resources for minor purposes.This paper presents a"making concessions in order to gain advantages"algorithm—an improved algorithm based on ACO algorithm for job scheduling problem. With a matching factor, the improved ACO is applied to heterogeneous clusters. Experimental results show that improved ACO approach can solve the problem, improve the average job response time and CPU utilization.