Cluster Load Estimation for Stateless Schedulers in Datacenters

Abstract

In probe-based distributed schedulers, little information is known about the state of the cluster. As a result, there is uncertainty about the underlying resource demand and usage. To efficiently leverage cloud datacenters' resources while maintaining the expected performance, one must address the question of how to achieve a good and accurate estimation of the cluster utilization in a stateless manner. We propose a scalable and efficient algorithm to estimate cluster load with a predetermined margin of error and confidence level. This algorithm can be used by cloud service providers to improve resource management systems and to estimate resource utilization. Due to its simplicity, the algorithm can be used in probe-based schedulers such as Sparrow, Tarcil, Piper, and Hawk.