Multi-criteria malleable task management for hybrid-cloud platforms

Abstract

The use of large distributed computing infrastructure is a means to address the ever increasing resource demands of scientific and commercial applications. The scale of current large-scale computing infrastructures and their heterogeneity make scheduling applications an increasingly complex task. Cloud computing minimises the heterogeneity by using virtualisation mechanisms, but poses new challenges to middleware developers, such as the management of virtualisation, elasticity and economic models. In this context, this work proposes algorithms for efficient scheduling and execution of malleable computing tasks with high granularity while taking into account multiple optimisation criteria such as resource cost and computation time. We focus on hybrid platforms that comprise both clusters and cloud providers. We define and formalise the main aspects of the problem, introduce the difference between local and global scheduling algorithms and evaluate their efficiency using discrete-event simulation.