Input-driven dynamic execution prediction of streaming applications

Abstract

Streaming applications are promising targets for effectively utilizing multicores because of their inherent amenability to pipelined parallelism. While existing methods of orchestrating streaming programs on multicores have mostly been static, real-world applications show ample variations in execution time that may cause the achieved speedup and throughput to be sub-optimal. One of the principle challenges for moving towards dynamic pipeline balancing has been the lack of approaches that can predict upcoming dynamic variations in execution efficiently, well before they occur.In this paper, we propose an automated dynamic execution behavior prediction approach that can be used to efficiently estimate the time that will be spent in different pipeline stages for upcoming inputs without requiring program execution. This enables dynamic balancing or scheduling of execution to achieve better speedup. Our approach first uses dynamic taint analysis to automatically generate an input-based execution characterization of the streaming program, which identifies the key control points where variation in execution might occur with the associated input elements that cause these variations. We then automatically generate a lightweight emulator from the program using this characterization that can predict the execution paths taken for new streaming inputs and provide an execution time estimates and possible dynamic variations. We present experimental evidence that our technique can accurately and efficiently estimate execution behaviors for several benchmarks. Our experiments show that dynamic pipeline balancing using our predicted execution behavior can achieve considerably higher speedup than static balancing.

FlipTracker: Understanding Natural Error Resilience in HPC Applications

“…ACL Table Generation. The algorithm to generate an ACL table is motivated by dynamic taint analysis in the security research [30], [31], [32], which focuses on computations affected by contaminated sources. The difference between taint analysis and our approach is that we exclude tainted locations that are never used as well as those that are overwritten by an uncorrupted value from the untainted location set.…”