A QoS Diagnosis Methodology Based on Dependency Matrix with Reputation Factor in SOA

This master thesis presents a novel approach to detect multiple faulty services by probing service information in large business processes. Our goal is to monitor running business processes, detect QoS violations, and diagnose where the root causes are. A dependency matrix is generated before runtime, and this matrix is used to help find the faulty services during runtime. Before business processes are executed, our system analyzes business process structures and service response times. The system will then perform evidence channel selection based on the above and record the data into a database. When business processes are running, the monitoring unit will check whether these business processes' end-to-end response times exceed their thresholds. If the monitoring unit finds an end-to-end response time violation, a diagnosis unit will be triggered to find where the faulty services are according to the information previously logged in the database. In our system, monitoring agents periodically probe information from part of the services. A dependency matrix contains the causal relationship between probes and services. In order to improve diagnosis performance, we introduce predicate of probes (PoP) to fully utilize the collected information. I also introduce a reputation factor to reduce the false positive rate. Our contributions can be summarized as follows: (1) It is not like probability-based diagnosis, model-based, and case-based methodologies, which require historical data and expert knowledge. (2) It can identify multiple faulty services in one round. (3) It compares data between probes to fully utilize collected information. (4) It introduces the reputation factor to help reduce the false positive rate. My main contribution includes (1) implementing evidence channel selection and diagnosis runtime support on LLAMA middleware, (2) co-inventing predicate of probes (PoP) to improve the completeness rate, (3) introducing a reputation factor to reduce the false positive rate