Workflow scheduling on virtualized servers

Abstract

Workflow applications comprise a number of structured tasks and computations featuring application services to be executed and the dependencies between these services. This paper deals with the problem of scheduling workflow applications, represented by directed acyclic graphs, on a set of virtualized servers. Each server hosts multiple virtual machines. Virtual machines sharing a host can communicate with each other, and with virtual machines hosted in different servers. The aim is to partition the application services and distribute each partition among the virtual machines in such a way that the dependencies are respected, the response time is minimized improving the quality of service and the intra-and inter-virtual machine communications are minimized. We model this problem as a workflow scheduling problem with hierarchical communications. The main contribution is to provide an evolutionarybased scheduling algorithm that considers this model when scheduling the applications. Simulation results demonstrate the effectiveness of the provided algorithm when compared with a related approach on a set of real-world applications emphasizing the interest of the approach.