A Survey of Fundamental Principles and Analysis for Job Scheduling on Warehouse-Scale Computers

Abstract

Over the last ten years, the search for efficient scheduling algorithms that can cope with heterogeneous workload demands on large (warehouse) scale computers has reached a feverish tempo. We focus on examining scheduling techniques for highly parallelizable jobs on warehouse-scale computers through the lens of basic results in relevant fundamental theories. The objective of this survey is to connect the disparate scheduling ideas and approaches in the literature under a loose framework of mathematical results that can be used to compare superficially different scheduling methodologies under a common goal. As the mathematical problem is NP-Hard in general, we do not emphasize rigorous mathematical proof, rather, we advocate for the use of mathematical results to guide intuition. We provide readers with some basic tools to use in navigating the fragmented research around job scheduling for distributed applications. We also highlight some common misunderstandings of fundamental theory in the literature that are skewing results and may be limiting research progress.