Qos accountability management in service oriented architectures
Kwei-Jay Lin, Yue Zhang · 2008
In this dissertation, a QoS accountability management model is designed to detect, diagnose, and defuse the root cause of problems when service delivery exceptions (such as incorrect result or SLA violation) occur in Service-Oriented Architectures (SOA). Although SOA provides a powerful and flexible paradigm to compose dynamic service processes using individual atomic services, it is necessary to manage end-to-end QoS at runtime for performance assurance. This accountable service model adopts the Bayesian networks to identify the most likely problematic services and inspects the details of a service execution when the service may be the root cause of a problem. A hierarchical service accountability diagnosis mechanism is also designed to achieve an efficient and scalable mechanism to pinpoint the service(s) responsible for failures. Several optimization algorithms, including evidence channel selection algorithms and agent selection algorithms, are designed to optimize the management framework cost and performance. Evidence channel selection algorithms are designed to specify which services in a business process should be monitored to achieve low data collection cost and high diagnosis correctness. They are modeled as the classic facilities location problems: k-median, Set-Covering, and Uncapacitated Facility Location (UFL). Accountability framework also deploys software monitoring agents. Agent selection algorithms are designed to integrate service selection with agent selection to minimize monitoring infrastructure cost. One is called IGA which is able to achieve efficient service selection and bounded agent cost by modeling the problem as the Weighted Set Covering (WSC) problem. Another heuristic algorithm is also designed to estimate the agent cost as part of the service utility function. This dissertation also presents the architecture and implementation of LLAMA middleware. It supports SOA-based service process composition, run-time management, and configuration. LLAMA is integrated into the Mule service deployment platform and leverages its existing monitoring capabilities. Instances of LLAMA's remote accountability agents can be deployed by providers specifications to guarantee their service's performance. These agents in turn allow LLAMA's accountability authority (AA) to diagnose process problems and apply any necessary reconfiguration measures.