Efficiently Speeding up Sequential Computation through the N-Way Programming Model

摘要

With core counts on the rise, the sequential components of applications are becoming the major bottleneck in performance scaling as predicted by Amdahl's law. We are therefore faced with the simultaneous problems of occupying an increasing number of cores and speeding up sequential sections. In this work, we reconcile these two seemingly incompatible problems with a novel programming model called N-way. The core idea behind N-way is to benefit from the algorithmic diversity available to express certain key computational steps. By simultaneously launching in parallel multiple ways to solve a given computation, a runtime can just-in-time pick the best (for example the fastest) way and therefore achieve speedup. Previous work has demonstrated the benefits of such an approach but has not addressed its inherent waste. In this work, we focus on providing a mathematically sound learning-based statistical model that can be used by a runtime to determine the optimal balance between resources used and benefits obtainable through N-way. We further describe a dynamic culling mechanism to further reduce resource waste. We present abstractions and a runtime support to cleanly encapsulate the computational-options and monitor their progress. We demonstrate a low-overhead runtime that achieves significant speedup over a range of widely used kernels. Our results demonstrate super-linear speedups in certain cases.