Lightweight Task Analysis for Cache-Aware Scheduling on Heterogeneous Clusters

TL;DR

This paper introduces a fast, accurate method for predicting program cache stress using minimal memory access observations, aiding task scheduling on heterogeneous clusters.

Contribution

It presents a novel cache stress characterization based on stack distance distributions that enables quick performance prediction for scheduling.

Findings

Accurate cache stress prediction with minimal data

Constant-time performance estimation

Effective for heterogeneous cluster scheduling

Abstract

We present a novel characterization of how a program stresses cache. This characterization permits fast performance prediction in order to simulate and assist task scheduling on heterogeneous clusters. It is based on the estimation of stack distance probability distributions. The analysis requires the observation of a very small subset of memory accesses, and yields a reasonable to very accurate prediction in constant time.