The Research On Multi-Objective Task Scheduling Strategy In Peer-to-Peer Environment

With peer-to-peer computing model rising, the substantial increase in network bandwidth and Internet computing power increasing rapidly, how to make full use of these network resources, to construct a large-scale, highly scalable, highly reliable, high-performance distributed computing system in a dynamic peer-to-peer network environment, which research is a hot one in recent years. Task scheduling is a key technology in P2P computing, it is a direct impact on the computing performance of the entire system. Task scheduling problem is that assigning a group of parallel processing tasks to nodes in accordance with the timing regulations by certain scheduling strategy, with a view to obtaining a better system performance.A task scheduling algorithm based on similarity is proposed to achieve load balancing and make good use of the network resources. To determine the nodes to be scheduled, the way of computing similarity between the current task and the past task is adopted. To get the nodes set, the measure of computing the similarity between the source requirements of tasks and the sources that nodes supply is taken. On this basis, the algorithm of task scheduling is given. Experiment results show that the proposed algorithm can efficiently reduce the "BUMP" phenomenon, achieve load balancing and make use of the network resources.The features of the P2P network make its task scheduling performance constrain by multi-factor. Considering execution time, communication time among nodes and task scheduling costs, a parallel task scheduling algorithm with multi-objective constraints is presented. The mathematical model of the multi-objective task scheduling is first proposed which defines the requirement relation matrix for each objective by using the relations between the task requirement and the nodes performance. Then all the relation matrices are transformed into fuzzy matrix by using membership functions. The proportions determined by the different effect on the final decision of each objective are applied to convert multi-objective task scheduling problem to that of the single one, and meanwhile the solution of best distribution on n tasks and m nodes is carried out by adopting Hungary algorithm. Experimental results show that the task scheduling algorithm with multi-objective constraints has better performances than the traditional methods.The model on task scheduling division is proposed to satisfy users' qulity service need. Firstly, the tasks are divided into real-time tasks and non-real time tasks according to the status of task requirements. Then two different task scheduling models based on the scheduling characteristics and existing task types are depicted by taking advantage of queuing theory, and also the rules of queuing and task scheduling are given. The experiments are used to analyze the effect of the task scheduling model proposed here in the existing algorithms, and which show the validity of classifying tasks and importing queuing theory in the task scheduling.