Research On Dynamic Task Scheduling Strategy Based On Resource-matching Rate In Peer-to-Peer Network

With the development of network technique and appearance of problems in the traditional C/S architecture, Peer-To-Peer (P2P) and other distributed technology become the research focus gradually in the recent years. Different from the centralized traditional architecture, P2P makes function of the server scattered throughout the whole network by marginalization strategy, which makes each peer in the network playing the role of client and server at the same time and perfectly solves the problem of "single point failure" and the overload of server in the traditional architecture. In addition, P2P has some other features, such as the freedom of entering into and leaving from the network, good scalability and so on.As we all know, it's the key problem that how to schedule the tasks to peers spreading through the network when doing high performance computing in P2P network. But lots of new problems of task scheduling are introduced because of the new features of P2P, like the impact on scheduling performance caused by the dynamic, etc al. This paper does research on the problems of task scheduling in P2P network, and the main results are shown as follows:Firstly, a pair of protocol called Request and Response is designed to obtain the resource information of peers as well as to estimate the transmission delay between peers in the target network. In unstructured P2P network, the nodes are organized together as a random graph. As a result, the performance of nodes differs a lot, and it's hard to estimate the transmission delay between them. But the performance of peers and transmission delay impact the performance of scheduling algorithm a lot. Preparing for task scheduling, the Request and response protocol are designed to obtain the performance information of peers that can be scheduled, as well as to estimate the transmission delay.Secondly, the Check Point technique was imported to relieve the impact on scheduling performance caused by the dynamic. In the process of task executing, the node periodically saves the immediate result. So when re-executing the task because of node leaving, the immediate result saved before can be used to resume the task which relieves the impact caused by the dynamic and shortens the executing time.Finally, a scheduling strategy based on resource matching rate is proposed. In the scheduling process, the nodes of peer class with maximum matching rate are considered at first. If failed, the peer classes which outperform that peer class are considered. Besides, the task resuming algorithm is proposed which runs when the peer executing tasks leaves from network. The experimental results show that the algorithm proposed outperforms HNPD,FCBSH,MPCGSR and TSIA algorithms in terms of execution time. Besides, the higher the leaving rate is, the better performance the algorithm shows.