Automated fault-injection-based dependability analysis of distributed computer systems

Abstract Recently, there has been interest in developing a dependability benchmarks for computer systems. This will require a way to inject several different types of faults into many,different platforms and a way to collect and compare the results. Analyzing complex heterogeneous distributed systems share the same needs. The current approach to building fault injection tool is inappropriate for these goals. NFTAPE (Networked Fault Tolerance and Performance Evaluator), a tool for designing and executing fault injection campaigns, make these goals possible by using lightweight fault injectors (LWFI). NFTAPE provides (1) an environment for executing fault injection campaigns using a library of fault injection programs (such as LWFI, triggers, monitors, and so on), (2) support services for communicating between test nodes, managing processes, and logging results, and (3) a programming model and API for writing support processes including exist fault injectors. Unlike older fault injectors, NFTAPE can use any fault injection method (software implementedfault injection, physical fault injection, or simulation). A few new variations on fault injections methods have come out developing NFTAPE so far: driver-based fault injector, target-specific fault injectors and performance faults. Already, NFTAPE has been used to run fault injection campaigns on several platforms (Linux, Solaris, LynxOS, and Windows) using several fault injection methods (debugger-based, driver-based, target-specific, and special-purpose hardware-based).