Static Analysis for Checkpoint Size Reduction in Array-Based Programs

As modern supercomputing systems reach the peta-flop performance range, they grow in both size and complexity. This makes them increasingly vulnerable to failures from a variety of causes. Checkpointing is a popular technique for tolerating such failures, enabling applications to periodically save their state and restart computation after a failure. Although a variety of automated system-level checkpointing solutions are currently available to High Performance Computing users, manual application-level checkpointing remains more popular due to its superior performance. This paper presents a compiler analysis that improves the performance of automated checkpointing by eliminating dead state, which will be overwritten before it is read by the application. In particular, the analysis focuses on 1-dimensional arrays that are accessed in loops with bounds that may be finite, linear or computed during the loop itself using linear expressions. As a concrete example we show how the analysis enables a 45% reduction in checkpoint size of mdcask, one of the ASCI Purple benchmarks [1].