Research For Scheduling Algorithm Based On Load Balancing In Distributed Heterogeneous Systems

With the social development and economic advances, the prediction of earthquake disaster is more and more important for promotion of social harmony. With the rapid development of modern computer technology, calculations of a growing number of projects are dependent on large-scale high-performance supercomputers to solve, such as the earthquake disaster prediction. And engineering calculations also drive the development of high-performance computers. Currently, the parallel system, grid computing and other distributed technologys are increasingly becoming important approaches to solve the engineering problems. In order to improve the performance of distributed systems, the scheduling have became a heated issue in morden time. Various scheduling algorithms have been proposed by a large number of researchers. They have different purposes, but in general are intended to improve the performance of distributed systems. In this paper, we analyse the tasks of earthquake disaster engineering in heterogeneous distributed computing system, and thoroughly studied the problem of task scheduling in order to improve the performance of the load balancing in heterogeneous distributed system.Firstly, in order to solve the problem of load balancing in distributed systems, we designed an improved dynamic genetic algorithm named IDGA (Improved Dynamic Algorithm) as the scheduling strategy of a heterogeneous distributed system. Taking into account the traditional genetic algorithm need to set the maximum evolution generation as a const, IDGA algorithm which overcomes the defects of the classic GA can dynamically change the setting of the maximum evolution generation. The IDGA algorithm tries to remember the optimal chrome during the procedure of all over the evolution.Secondly, because the IDGA algorithm can dynamically set the maximum evolutionary generation, the genetic algorithm may cause an increased time complexity. In order to improve the efficiency of the IDGA algorithm, we proposed a model which based on the MapReduce. Our target is to parallel the IDGA algorithm in our proposed model.Finally, we proposed a data structure-a binary choice tree to organize the initial population of the genetic algorithm. Using this structure, the time complexity of the selection operation in the genetic algorithm will be improved. In detail, the binary choice tree is a balancing tree, so it can improve time complexity of the selection operation from O (n) to O (log n), which n representative scale of the populations. Each node of the binary tree must remembers three fitnesses, those are:the fitness of right subtree, the fitness of left subtree and the fintness of itself. Therefore, before we have to deside which one would be selected,right subtree, left subtree of the curent node, we can produce an random number N, which N is less than1and bigger than0, to help us to make the decision. So the binary choice tree significantly improves the efficiency of selection operation.