Fuzzy Clustering Of Grid Resources Based On Mobile Agent

In grid computing environment, resources are heterogeneous, dynamic and self-government, so it is required that appropriate resource can be cho-sen efficiently in resource allocation and task scheduling. Thus the pre-treatment and analysis of resources are rather important. Fuzzy clustering analysis is a kind of effective grid resources pretreatment method. Currently, center fuzzy clustering method used for grid resources is to collect all the available resource performance information, and large numbers of clustering calculation concentrated on global management node, so computing per-formance of the global management node has become the bottleneck. Fuzzy clustering costs less time, when the amount of resources is small. In grid computing environment, the amount of resources'is enormous, which will make the scale of grid resources clustering computing to increase rapidly, the time complexity of cluster analyzing process to become very large and the process to become low efficient.Grid and mobile agent are both promising distributed computing tech-niques. Some researches have combined mobile agent with grid. Hence it is feasible to improve grid resources'fuzzy clustering method into a parallel fuzzy clustering method with the use of mobile agent technology. This essay presents a grid resource fuzzy clustering model based on mobile agent which can be regarded as a sub-module of the mobile agent based grid resource management model, that is, the fuzzy clustering task concentrated on global management node will be decomposed into some sub-tasks which can be parallel processed, with the use of grid virtual or-ganization structure, with the mobile agent as the middleware.A model of grid resource fuzzy clustering based on mobile agent in-cludes monitoring module and fuzzy clustering module. Monitoring module is consisted of resource performance Sensor Agent (SA) and Virtual Or-ganization Performance Management Agent (VOMA), whose main function is to collect the performance data of resources, to provide the relative data to fuzzy clustering module. Fuzzy clustering module, consisted of Virtual Or-ganization Clustering Agent (VOCA), Global Clustering Agent (GCA), Data Transform Agent (DTA) and Communication Agent (CA), is responsi- ble for calculating the fuzzy equivalent matrix with the original data matrix. Fuzzy clustering module is the most important part of this essay.This essay chooses a java-based Aglet platform as the development platform of mobile agent. The grid resource fuzzy clustering is implemented based on Mobile Agent. In Aglet platform, the fuzzy clustering process is as follow:1 SA collects the performance information of resources, and VOMA gets the performance data of local virtual organization resources.2 VOCA calculates the minimum and maximum groups of the local data in every dimension, while GCA calculates the global minimum and maximum groups.3 VOCA processes the normalization of the local data.4 VOCA does the parallel calculation of the sub-blocks of the fuzzy similarity matrix.5 DTA migrates the sub-blocks of the fuzzy similarity matrix, and then VOCA gets the global fuzzy similar matrix.6 VOCA does the parallel calculation of the fuzzy equivalent matrix.7 DTA migrates the sub-block of the fuzzy equivalent matrix.8 VOCA determines whether the calculation of the fuzzy equivalent ma-trix has been completed.9 GCA calculates the fuzzy clustering result, and reports the result. Altogether, compared with the center grid resource fuzzy clustering method currently used, the major improvements of grid resource fuzzy clus-tering method based on mobile agent (GRFCMA) are:1 It can reduce the time that fuzzy clustering costs and improve the effi-ciency of fuzzy clustering through the use of parallel calculation of fuzzy similar matrix and the fuzzy equivalent matrix by mobile agents with B-shift algorithm.2 The amount of data storage in the global management will be greatly reduced. Global management node doesn't need to store the perform-ance data of all the resources, and the key data will be stored in the lo-cal management nodes of the virtual organization.3 It makes the fuzzy clustering task on the global management node to be shared by the various virtual organizations, makes full use of the avail-able nodes in the grid environment, and reduces the work pressure of the global management node. Since the global management node will not be overloaded, and other nodes won't be too free, it will balance Grid workloads to a certain extent.This essay designs several simulation experiments that contrast the mobile agent based grid resource fuzzy clustering method with that of center fuzzy clustering. In the simulation experimental environment, it is assumed that one machine was Global Management Node and each of the other ma-chines was a Virtual Organization Management Node. The time complexity of fuzzy clustering is related to the amount of the resources and the number of the performance dimension, so the experiments are designed as follow: the amount of resources and the number of performance dimension are set as two parameters, and then one of the two parameters was fixed, while the other parameter increases gradually. Thus, two groups of contrast experi-ments under the same experimental conditions are set: the simulation ex-periment of changing the amount of resources and the simulation experi-ment of changing the number of the data dimension. The input data of the simulation experiment is the document for storing the original data matrix, and the original data is chosen randomly from the data source of http://kdd.ics.uci.edu. The output result should be the fuzzy equivalent ma-trix.The result of simulation experiments shows that: regardless of the in-creasing the resources'amount or the increasing of the performance dimen-sions'number, the time that the mobile agent based fuzzy clustering method costs, is shorter than the time that center fuzzy clustering method costs. With the increasing of the clustering calculation scale, the time advantage of the method based on mobile agent will be more obvious. The fuzzy cluster-ing method based on mobile agent is suitable for heterogeneous and dy-namic grid environment.The next phase of this work will be how to get the highly accurate forecasting performance value of the resources with mobile agents, then to analyze the predictive values of fuzzy clustering and how to use the result of work for the resource allocation, to choose the suitable resources for the ap-plication and the task according to the requests.