Improving service level agreements for a job scheduler by visualizing simulations

Abstract

Currently, job owners at Google do not have a good way to generate suitable Service Level Agreements (SLAs), which means that they cannot accurately communicate their intentions to the job scheduler. This means that the owner's job might not finish on time or at all. The solution described in this thesis helps users visualize design changes to SLAs and use simulation to explore the behavior resulting from the SLAs. I have designed and begun development of a visualization and simulation framework that allows users to see how the job scheduler's behavior might vary under different SLA parameters. This thesis describes the steps made towards designing and implementing a system that both helps users visualize SLAs and their reward functions, and allows users to create an SLA and gain an idea of the behavior of a job scheduler with the SLA as input.