Grid Performance Prediction Based On RBF Neural Network

Based on the conception of Electronic Grid, Grid computing was proposed and developed rapidly with internet technology. In a Grid computing schema, various high-powered computing resources, storage, data and other specific resources, which are geographically distributed and heterogeneous, are combined into a dynamically optimal composition of these resources via internet, and establish a collaborative high-performance computing to solve some important application problems. As one kind of distributed computing technology, it is also known as wide-area high-performance meta computing technology. With the shielding of distinctions between hardwares, operation systems, organizations and regions by Grid infrastructure, a"Virtual Super Computer", which is a combination of all kinds of resources, can be shared and managed transparently and efficiently by different users and organizations. Grid computing has powerful data processing performance and is capable of making full use of idle computing resources in internet.In the process of solving large-scale problems through Grid resources, due to the dynamic changes of natural resources, computational Grid needs to have these real-time, accurate and dynamic performance information to predict the Grid performance, and provides the users of the Grid resources with the job scheduling basis. The importance of effective monitoring and predicting to Grid resources is that there will be a real-time reaction to these dynamic information, such information can provide the basis to the more reasonable applications of Grid resources, and the improvement of the Grid application performance.In the background of the project"Real-time monitoring and predicting of Grid applications'execution performance"which is funded by National Natural Science Foundation, this paper established a set of methods and models to monitor the execution performance of Grid applications and provide a real-time predicting. The trend of Grid resources performance can be concluded from the dynamic information of measurement and predicting, to improve the execution performance of applications.This paper stated the grid architecture in the part of monitoring, that is, early five hourglass structure, OGSA (Open Grid Services Architecture) and WSRF (Web services resource framework), and the two important support technical in OGSA are Web Service and Globus Toolkit. In this paper, the monitoring data is obtained from some services of Globus Toolkit in the basis of Web Service. The major one of these services is Grid Resource Management (GRAM), and it is a functional module in GT. As the component of the mandate infrastructure s, GRAM provides a set of Web services of submission, monitoring and cancellation in Grid computing resources. After allocating the operation to the designated Grid resources, GRAM will has an operating ID. By using the ID, users can operate the operations in the life cycle, access to the information of system resources'usage, and then the information transfer between each grid node can be processed through reliable file transfer service ( RFT).The program running in Grid, no matter running in one node or multi-nodes, will be finally allocated to Grid Resources with the form of process. Consequently, the Grid Process Monitoring should contain two parts. That is, the monitoring to the allocation of the Grid Program and the monitoring to the grid process in every node. The first part should get information that which node the grid process is on earth dispatched to; the second part should monitor the information (CPU usage, Memory usage, Net IO rate-if necessary) of the grid process in every node.In the prediction part, this paper mainly introduced RBF prediction algorithm in Artificial Neural Networks. RBF network is one kind of feed-forward Artificial Neural Networks algorithm, and has the characteristic of strong approximation and the input-output mapping function, and its learning process has faster convergence speed. In RBF, there are three main system parameters need to be determined, which are the hidden layer-function center, variance and the weight of hidden layer element to output unit. The choosing of these parameters will affect the RBF network prediction directly. Therefore, in the design of the system, main parameters and data which need to modify are typed into the files in order to facilitate the process of prediction experiment. So through the multiple parameter adjustment of the same group of prediction data, and the comparison of error, user can choose a set of network configuration parameters which have the best prediction effect, and predict the resources and the application programs in Grid by using these parameters in RBF, and the prediction results can be obtained as the evaluation criteria of a Grid.Finally through experiments, this paper accessed to the prediction value and compared with the real value in graph, and found that the RBF network can be used in the prediction of the resources and the application programs in Grid. Although the experimental data in the prediction have fluctuations, but its image's fluctuations are in the general vicinity of real value, it means that the data from prediction has the same trend with the real value, so the algorithm has certain research value in the prediction of the resources and the application programs in Grid.In the future work, there's the need to evaluate and classify the performance of Grid node by using of the existing monitoring and forecasting data. Adding the monitoring of Web services into the process monitoring, it is not only to be able to monitor all application programs in Grid, but also the services of some components. In the aspect of algorithm, feedback algorithm in Artificial Neural Networks, such as Hopfield algorithm, and some prediction algorithm in other areas can be applied to the Grid performance prediction.As an emerging technology, Grid is mainly in the theoretical and experimental stage currently, and its correlative standards and norms are still in need of establishment and improvement. The monitoring and forecasting methods in Grid will definitely become the important problem in the research of Grid, which is why we also need to be more applied research, thus we need more applied research and to find more valuable ways in which the technology can be used in practice as soon as possible.