Online reliability computing of composite services based on program invariants

Reliability is an essential software quality requirement, especially for online service software. Without an accurate prediction of service reliability, any unexpected failure can disrupt service. The majority of existing models use static data collected prior to the release of the software. These types of models may predict the reliability of the software as it was during the data collection phase. However, online service software is continuously evolving, and their behaviors can be changed by the runtime usage. Thus, the prediction made by static data can be inaccurate. We present an approach to tackle this challenge by taking into account software runtime behavior in our reliability prediction. We used a data mining tool, Daikon, to collect likely invariants of the software to capture its states in the runtime. This runtime information is then used to compute the reliability of the software by using our port-based reliability model.