Research On Resource Allocation Algorithm Based On Collective Interests In Distributed Environment

With the rapid development of science and information technology, the computer system has been transformed from the stand-alone systems to the distributed systems. In the distributed environment, how to improve the computing and service capabilities by means of the integration of different resources through network, has been widely focused on. In this paper, a resource agent-based grid platform is designed and implemented, and resource allocation of two typical distributed environments are investigated.In the distributed environment, user applications usually are divided into multiple independent and related task. There are execution order and transmission of data between the related tasks. In this paper, the two model of complex resources allocation in the case of independent and related tasks are built up by game theory respectively, and the existence of Nash equilibrium is proved. Furthermore, the influence of the game between all tasks on the whole system performance will be analyzed. In the resource allocation model of independent task, the collective interests can be reflected by the total cost of all tasks. It is an important indicator of the system performance. In this study, a resource allocation algorithm (RACGR) is proposed based on the maximizing collective interests by changing the payoff function of the players in the game. In the resource allocation model of the related tasks, the relations between all tasks can be represented with a DAG map, and collective interests can be regarded as the makespan. Makespan is the time from beginning of the first task to the ending of the last task. We present another resource allocation algorithm (TSBS) based on the topological sorting. It can reduce the influence of static games to the makespan via the sorting of all tasks. As a result, the makespan is shortened and the system performance is improved.In simulation experiment, RACGR is compared with Nash equilibrium and RR based on independent task, and TSBS is compared with Max-min and Min-min based on related task (DAG map). We analysis the influence of the spend function to the number of suntasks on each resource. Finally, we can get the conclusion that RACGR algorithm has better performance on the total system cost and fairness of resource, while TSBS algorithm can effectively improve resource utilization and shorten the makespan of random DAG map and DAG map which is used in the real world.