Locating failure-inducing environment changes

Abstract

Traditionally, debugging refers to the process of locating the program portions which are responsible for a program failure. However, a program also fails when the execution environment does not meet the requirement/assumption of the program. Unfortunately, few existing debugging techniques addresses the problem of changing operating system environment. In this paper, we propose an effective record-replay technique called Semi-replay to solve this problem. Semi-replay records all the essential interactions between an application and its underlying operating system environment where it successfully executed. Semi-replay then allows the recorded interactions to be partially replayed and partially executed in another operating system to identify those interactions which contribute to the root cause of the application failure induced by the environment changes. We have conducted three case studies on real-life programs which show the significance and efficiency of the Semi-replay technique in locating failure-inducing environment changes. We have also implemented a tool for the Linux kernel to demonstrate the feasibility of the proposed approach.