The Research On Information Collection And Service Discovery In Grid Monitoring System

Grid monitoring system aims at collecting real time status information of resources and functional components, in order to make sure the performance of grid. But there are lots of problems in the existing systems, such as monitoring data included is not comprehensive, no unified collecting structure is defined, and service discovery is not efficient. As a result, the research on Information Collection and Service Discovery are very important. This thesis focuses on these two mechanisms of ChinaGrid SuperVison.Information Collection includes information collection architecture, communication protocol and metrics definition. The mechanism inherits the advantages of some popular distributed monitoring systems, and develops their Information Collection modules to a system that collects comprehensive information in a unified way, and reduces the collecting load. Also it makes the definitions of different metrics more flexible and standard.Considering the application-oriented characteristic of ChinaGrid, a hybrid Peer-to-Peer overlay named Peer-Tree is built to enable efficient service discovery. Registry services representing organizations are organized in two layers named Peer Layer and Tree Layer. Join Peer-Tree and Leave Peer-Tree Algorithm enable the organizations with similar service interests hierarchically structured in a tree, while the roots of those trees distribute in Peer Layer. Learn Neighbor Mechanism not only tells organizations to choose the best neighbors to forward the service queries, but also changes Peer Layer into a small world overlay; it improves the service efficiency.After functional and performance tests, it can be found that Information Collection Mechanism of CGSV is a successful design, while Service Discovery Mechanism verifies its Peer Layer's small world properties by simulations. The tests also reveal that there exists a most efficient discovery when certain node distribution between two layers is specified, which testifies that this kind of Peer-Tree is better than both fully distributed system and fully hierarchical one.