Research On Several Key Technologies In New Generation Internet Service And Failure Detection

Services namagement is one of the core issues for new generation of Internet service. Internet services will not be controllable in future if there is no scientific method to manage services. How to formally describe internal structure and interface of service, how to assemble more advanced services, how to find and match service on demand, how to describe the properties and performance metrics of failure detection service, how to effectively deploy failure detection services, how to evaluate the performance of failure detection service, how to improve the performance of failure detection services, all these theoretical issues must be resolved.To address above critical scientific problems, this dissertation mainly focuses on several key technologies of new generation Internet service and fault detection with help of theoretical analysis and experimental simulations. The main contributions are summarized as follow:1) We analyze the characteristics of the heterogeneous network infrastructure, personalized user requirements, many participants for new generation Internet. We propose three service models based on the views of participators, semantic description and implementation. These models provide uniform, effective, intelligent mechanisms to manage, describe, implement NGI service. A hierarchical architecture is introduced to direct further NGI service management research。2) We first analyze the various requirements for Grid failure detection services, formal description of the failure model, failure detector model as well as performance evaluation metrics. Then propose multi-level failure detectors which can implement different failure detection mechanisms based on different Grid environment (nodes, the sites, virtual organizations) to detect the failures of process, host, site respectively and guarantee all Grid requirements.3) Interoperability is another important property which must be guaranteed for failure detection in Grid. We use ontology to describe failure detection service and propose a service-oriented architecture framework to make it to be independent of the specific implementation and guarantee its interoperability.4) We do the comprehensive simulations to investigate the performance of heartbeat-style failure detector over proactive and reactive routing protocols with simulator. The performance different are analyzed using varying the number of nodes, mobility speed, traffic load, transmission range. Simulation shows that proactive performed better in average delay and average failure detection time, reactive performed better in normalized routing overhead, delivery ratio, energy consumption ratio, false detection ratio. The observations can be used to motivate and improve future implementations of failure detection service in Ad hoc wireless networks.5) Internet dynamic characters make it very difficult to understand message behavior and accurately predict heartbeat arrival time. To overcome this problem, a novel black-box model is proposed to predict next heartbeat arrival time. Heartbeat arrival time is modeled as Auto-Regressive process, heartbeat sent time is modeled as exogenous variable, the coefficients are estimated based on the sliding window of observations and this result is used to predict next heartbeat arrival time. Simulation shows this adaptive Auto-Regressive exogenous (ARX) model can accurately capture heartbeat arrival dynamics and make minimum prediction error under different network environments.6) Two-layer feed forward neural network is proposed to learn nonlinear characters of heartbeat messages, perform one-step-ahead prediction to estimate future heartbeat message delay. Inputs are one moving window of observations of the heartbeat delays, output is the one-step-ahead future value, the neural network is trained by back-propagation algorithm, its weights and basis are adjusted by approximate steepest descent rule. Simulation shows this adaptive algorithm can accurately capture heartbeat message dynamics and make better prediction result.7) We use a nonlinear autoregressive network with exogenous inputs to learn nonlinear and linear characters of heartbeat messages, perform one-step-ahead prediction to estimate future heartbeat delay. The inputs are two moving window observations of past heartbeat delays and heartbeat sending time, the output is next heartbeat delay, the network is trained by standard back-propagation algorithm, its weights and basis are adjusted by approximate steepest descent rule. Simulation shows this adaptive algorithm can accurately capture heartbeat dynamics over internet and make best prediction result.