FUD — Balancing Scheduling Parameters in Shared Computing Environments

摘要

Shared computing environments such as Cloud, HPC and Grid Computing present a challenge for scheduling systems as they seek to balance incoming requests with available resources, maintain high utilization, be fair among users, and cope with environmental dynamicity. In this paper, we will introduce the FUD theorem. The FUD theorem is based on the premise that a scheduler's desire to optimize the three system parameters: Fairness, Utilization and Dynamicity (FUD) comes at a cost. These parameters adversely affect one another, and thus, in a shared computing environment, a scheduler is unable to be fair while fully utilizing the available resources and still deal with the dynamicity of the environment. The presented FUD model argues for relaxing one of the three parameters to optimize scheduling decisions based on the remaining two.