(An) interaction-pattern-based approach to prevent performance degradation of fault detection in adaptive software

Abstract

A component-based approach is a promising technology to support dynamism in adaptive software. In this approach, many developers independently produce various components in their domains. During runtime, the components are dynamically integrated to handle dynamic situations and unexpected user requirements. Therefore, there are potential problems when heterogeneous components are integrated to support a user task. The adaptive software may face software failures caused by the nature of the adaptive software. The causes of the failure (called faults) are: limited resources (computational limitations, memory shortage), component interoperation mismatch, and internal errors of components. The message-sniffing-based monitoring fault detection is a popular method used to monitor black-box components and handle these types of faults during runtime in adaptive software. However, these methods have some performance problems of the target system because the fault detection process is continuously processing with the target system and consuming resources of the target system. There are three types of causes of the performance degradation: frequent monitoring, monitoring data transmission, and fault analysis overhead. This performance degradation has a serious effect on the timing of the target system, and may result in a software behavior that a correct service may deviate. In this paper, an interaction-pattern-based approach is proposed to prevent performance degradation of fault detection in adaptive software. The interaction-pattern-based approach can alleviate the performance degradation problem in detecting faults by monitoring/inspecting not all messages but the most important messages among message exchanges. To formalize message exchanges in fault detecting aspects, commonly observed message exchanges are analyzed and identified as interaction patterns. In addition, important messages to monitor/inspect are identified in the interaction patterns for the selective...