Grid Environment, Resource Discovery Mechanism

With the increase of the requirement of the high-performance, a single high-performance computer isn't competent to settle problems about the super large-scale. So geographically distributed and heterogeneous resources linked through a high speed network are needed to realize the overall sharing of the resources, including calculating resource, storing resource, information resources, knowledge resource and so on, dispel" detached island " of information, and solve the large-scale problem together.In the traditional single computer and airmada system, the distribution of resources is centralized. Before they are used, these resources are located fast and reliably and the influence of resource discovery to the performance of application is small. But in grid computing, because of wide distribute of resources and the existing the limit of bandwidth and delay and unreliability of network in current Internet, the discovery of resources in the wide area range will influence the computing performance to a great extent, therefore the power of resources discovery has determined use factor and friendly degree directly in grid.Firstly the thesis has discussed the characteristic and the development trend of the grid and analyzed some existing grid resources discovery models. Then the grid technology and the P2P technology are combined to propose one rely on un-centralized control, distributed, expand, self-organized, and meet the dynamic change of resource, and high-efficient resources discovery model. And this model combines C/S pattern with P2P pattern, not only various kinds of characteristics of grid resources are considered, but also the P2P characteristic and high-efficient resource search technology are utilized, thus fine realize the sharing of resource.Resource discovery includes the Key technologies about resource describe, storage and maintenance of resource information, resource query, and the spread of query message, etc. This thesis has discussed these technologies and researched them farther. In resources describe, in order to located remote resources effectively, we usepeer identification with only ID. In this way, peer ID can separate from its physical location as discovering; Peers can respond each other by define Advertisement. And in the model every Rendezvous Peer maintains those resources that published in it and other Peers that it knows. This embodies the decentralization and autonomy of maintenance. Resource query consider exponential law phenomenon (i.e. locality principle) in reality, so we adopt the improvement algorithm of flooding, the method of Multi-Cache (MC), which improved the search efficiency of resources. To guarantee the live, unsettled and non-circulate before dealing with intactly Using TTL life time parameter, duplicate test parameter, circulation parameter to control the spread of message.The design of the model is a complicated and big problem. Because of the reasons, for example time and objective practice environment, etc. In the thesis the realization involved the connections and discovery of peers. A lot of functions of the model still fail to realize, some technology still remains at theoretical research stage, remain to study and perfect further.