AntSM: Efficient Debugging for Shared Memory Parallel Programs

摘要: This paper describes AntSM, a system that uses the inherent parallelism of multi-threaded programs to reduce overhead statistical and invariant violations detection-based debugging tools. The runtime monitoring these tools leads high overheads. key insight AntSM is this can be reduced in parallel by performing sampled across regions program are similar actions. implements sampling using combination static dynamic analyses determine parts executing number threads those program. Experimental results, performed C-DIDUCE (a variant DIDUCE for C) tool on eleven Pthreads benchmarks from PARSEC suite, show up 18.14 times (and average 8.73 times) an eight-core machine relative naive port performs no sampling.