The Case for Resource Sharing in Scientific Workflow Executions

Scientic workows have become mainstream for conducting largescale scientic research. The execution of these applications can be very costly in terms of computational resources. Therefore, optimizing their resource utilization and efciency is highly desirable, even in computational environments where the processing resources are plentiful, such as clouds. In this work, we study the case of exploring shared multiprocessors within a single virtual machine. Using a public cloud provider and real-world applications, we show that the use of dedicated processors can lead to sub-optimal performance of scientic workows. This is a rst step towards the creation of a self-aware resource management system inline with the state-of-the-art multitenant platforms.