Highly scalable algorithms for scheduling tasks and provisioning machines on heterogeneous computing systems

Abstract

As high performance computing systems increase in size, new and more efficient algorithms are needed to schedule work on the machines, understand the performance trade-offs inherent in the system, and determine which machines to provision. The extreme scale of these newer systems requires unique task scheduling algorithms that are capable of handling millions of tasks and thousands of machines. A highly scalable scheduling algorithm is developed that computes high quality schedules, especially for large problem sizes. Large-scale computing systems also consume vast amounts of electricity, leading to high operating costs. Through the use of novel resource allocation techniques, system administrators can examine this trade-off space to quantify how much a given performance level will cost in electricity, or see what kind of performance can be expected when given an energy budget. Trading-off energy and makespan is often difficult for companies because it is unclear how each affects the profit. A monetary-based model of high performance computing is presented and a highly scalable algorithm is developed to quickly find the schedule that maximizes the profit per unit time. As more high performance computing needs are being met with cloud computing, algorithms are needed to determine the types of machines that are best suited to a particular workload. An algorithm is designed to find the best set of computing resources to allocate to the workload that takes into account the uncertainty in the task arrival rates, task execution times, and power consumption. Reward rate, cost, failure rate, and power consumption can be optimized, as desired, to optimally trade-off these conflicting objectives.