CATCH - Compiler-Assisted Techniques for Checkpointing

Abstract

A compiler-based approach to generating efficient checkpoints for process recovery is described. The presented approach to checkpointing is programmer, operating system, and hardware transparent. Compile-time information is exploited to maintain the desired checkpoint interval and to reduce the size of checkpoints. Compiler-generated sparse potential checkpoint code is used to maintain the desired checkpoint interval. Adaptive checkpointing has been developed to reduce the size of checkpoints by exploiting potentially large variations in memory usage. A training technique is used in selecting the low-cost, high-coverage potential checkpoints. Since the potential checkpoint selection problem is NP-complete, a heuristic algorithm has been developed to obtain a quick suboptimal solution. These compiler-assisted checkpointing techniques have been implemented in a modified version of the GNU C (GCC) compiler version of 1.34. Experiments utilizing the CATCH GCC compiler on SUN workstations are described. >