Event-based Analysis Framework For Distributed Software Behavior

With the rapid development of network technology, software systems have undergone profound changes in morphology. Distributed software systems have been widely used in various information fields, playing an irreplaceable role. How to recognize and enhance the credibility of such systems is an important guarantee of its service in the national key areas. Distributed software systems have the characteristics of open environment, huge magnitude and complex behavior, which are different from desktop software and embedded software. Because of these characteristics, traditional software engineering methods can not resolve all the credibility problems of distributed systems in the development stage. Therefore, during the execution of a distributed system, to monitor the executing status, to determine whether the behavior of real systems in line with expectations and then adjust whether the system healthy or not, is an effective way to improve the credibility of distributed system.There are many complex functional logic and collaborative interaction between nodes frequently in distributed software system, which result the run-time behaviors are complex and difficult to control. During the execution, a large number of fast-changing information state change events are generated. How to make use of these meaningful information state change events to analyze system behavior, and then to judge the health of the system, plays an important supporting role for enhancing the credibility of the distributed system. This paper designs and implements event-based framework for distributed system behavior analysis, and presents models and methods for distributed event aggregation, distributed system behavior specification description and distributed system behavior checking. The main work of this article includes the following aspects:1. In this paper, we present a method for distributed system behavior specification description based on complex event. We gather a series of basic events generated by distributed nodes, and constitute a complex event based the combinational logic and temporal logic between events, then describe the specification of the distributed system behavior which is related with this series of events.2. In order to solve the problem of distributed monitoring events gathering, this paper presents a method based on publish and subscribe mechanism. With this method, we choose low-invasive, lightweight method for probe injecting to collect the events from distributed nodes, and solve problems in the distributed event transmission by asynchronous, loosely coupled communication means, which can reduce congestion and ensure efficient delivery of events.3. This paper presents two mechanisms for checking the distributed system behavior: The on-line checking mechanism based on event stream, checks the behavior by processing complex events in the memory, so that ensure the timeliness of the procedure of behavior analyzing; The other checking mechanism based on historical information, implements by storing events in database and static query, which can provide a more adequate basis and improve the ability to distributed system behavior analysis.