Path-based Fault Injection

This paper presents a path-based fault-injection approach to increase the eeciency of computer system dependability evaluation. Our approach utilizes knowledge about the execution path to ensure that injected faults are activated. The control ow associated with each path is analyzed to determine the faults that would be activated by that path. By selecting a set of paths that covers a signiicant portion of the entire code space of the test program , a comprehensive fault set is constructed that is capable of directly aaecting all control-ow decision points. In contrast to this path-based approach, common fault-injection methods randomly select fault time and location parameters without considering the accompanying path and result in a lower level of fault activation. Although the path-based approach requires a pre-injection analysis, this time-cost is only incurred once and is amortized during the fault injection phase. 1 Introduction Fault injection is an important step in the process of validating and evaluating computer system dependability. Dependable systems incorporate fault-tolerant features to rst detect errors and then to mask, avoid, or recover from the eeects of those errors. Testing the dependability of these systems is equivalent to testing the functionality of the fault-tolerant mechanisms. Fault injection is essential to test these fault-tolerant mechanisms so that the system behavior can be studied a priori under realistic scenarios. Although the purpose of fault injection is to force the exercise of the fault-tolerant components of the target system , not all injected faults accomplish this purpose. Faults should not be placed where activation is impossible. For instance, faults in memory locations that are not assigned to an executing process will never be accessed and therefore never cause an error detection. In this paper, faults that are accessed are labeled as activated. Path-based injection is an approach that minimizes non-activated faults through an intelligent selection of fault parameters. Path-based injection is especially useful for testing an embedded system, which executes a single program repeatedly. For such a system, an involved eeort to thoroughly test the program with fault injection is jus-tiied. Path-based injection ensures that the entire fault