Transient Fault Tolerance via Dynamic Process-Level Redundancy

Transient faults are emerging as a critical concern in the reliability of microprocessors. While hardware reliability techniques are often employed for transient fault tolerance, software techniques represent a more cost-effective and flexible alternative. This paper proposes a software approach to transient fault tolerance which utilizes a run-time system to automatically apply process-level redundancy (PLR). PLR creates a set of redundant processes per application process and compares the processes during run time to guarantee correct execution. Redundancy at the process level allows the operating system to freely schedule the processes across all available hardware resources (i.e. extra threads or cores). PLR is a software-centric approach to transient fault tolerance in which the focus is shifted from ensuring correct hardware execution, to ensuring correct software execution. The software-centric approach is able to ignore many benign faults which do not propagate to affect the program output. In addition, the dynamic deployment creates a very flexible fault tolerant system which transparently applies PLR without prior modifications to the application, shared libraries, or operating system. Experiments using a real PLR prototype on an SMP machine demonstrate that PLR can effectively provide transient fault tolerance with a slowdown of only 1.26x.