Increasing System Availability with Local Recovery Based on Fault Localization

Due to the fact that software systems cannot be tested exhaustively, software systems must cope with residual defects at run-time. Local recovery is an approach for recovering from errors, in which only the defective parts of the system are recovered while the other parts are kept operational. To be efficient, local recovery must be aware of which component is at fault. In this paper, we combine a fault localization technique (spectrum-based fault localization, SFL) with local recovery techniques to achieve fully autonomous fault detection, isolation, and recovery. A framework is used for decomposing the system into separate units that can be recovered in isolation, while SFL is used for monitoring the activities of these units and diagnose the faulty one whenever an error is detected. We have applied our approach to MPlayer, a large open-source software. We have observed that SFL can increase the system availability by 23.4% on average.