TaskFlow: An Energy- and Makespan-Aware Task Placement Policy for Workflow Scheduling through Delay Management

Abstract

Datacenters need to become more power efficient for political and climate reasons. In this work, we introduce an idea for the community to further explore. We embed the idea in TaskFlow: a makespan conservative, energy-aware task placement policy for workflow scheduling. Using static, rough numbers and simulation, we obtain energy savings between [4.24, 47.00]% and [0.1, 13.6]%, respectively. We also present some pitfalls that should be investigated further, notably starvation of large tasks when using TaskFlow.