Performance and reliability analysis of the CORBA event and notification services

The Common Object Request Broker Architecture (CORBA), spear-headed by the Object Management Group (OMG), is an industry-wide standard that facilitate portable, distributed object-oriented computing. This dissertation focuses on the reliability and performance aspects of the CORBA services that facilitate asynchronous communication between CORBA objects. The dissertation first examine the drawbacks of the CORBA Event Service, and proposes an application-level recovery scheme based on state resynchronization to provide resiliency in the event of failures. The performance of the Event Service is then analyzed in detail using stochastic reward nets, a modeling formalism that extends stochastic Petri nets. The models allow performance analysis of the Event Service under bursts and non-bursty input conditions as well as under transient overload conditions. The dissertation then examines whether the CORBA Notification Service lends itself to reliable messaging. After identifying the delivery semantics provided, a framework for reliable messaging using the Notification Service is proposed. This is followed by models—based on queuing theory and Markov modeling theory—that can be used to analyze the performance of several QoS configurations of the Notification Service. The dissertation concludes by developing performability models to evaluate messaging services in distributed systems. This part also serves as an illustration to suggest how prevalent architecture-based techniques for software reliability and performance estimation call be extended to take into account specific architectural characteristics of applications.