Tightness results for malleable task scheduling algorithms

Abstract

Malleable tasks are a way of modelling jobs that can be parallelized to get a (usually sublinear) speedup. The best currently known approximation algorithms for scheduling malleable tasks with precedence constraints are a) 2.62-approximation for certain classes of precedence constraints such as series-parallel graphs [1], and b) 4.72-approximation for general graphs via linear programming [2]. We show that these rates are tight, i.e. there exist instances that achieve the upper bounds. © 2008 Springer-Verlag Berlin Heidelberg.