Algorithms to Improve Resource Utilization and Request Acceptance Rate in IaaS Cloud Scheduling

摘要

Computing infrastructure provisioning in Infrastructure as a Service (IaaS) cloud is done in the form of virtual machines.Haizea, a resource lease manager, provides four types of leases: Immediate, Best Effort (BE), Advance Reservation (AR) andDead Line Sensitive (DLS). AR leases are most privileged leases with “AR preempts other leases” policy, since they canpreempt & suspend other BE leases when demanded by consumers. This leads to two problems: 1) a set of BEs can gosuspended infinite number of time & 2) ARs, at the actual time of their resource allocation, can be rejected due to presence ofother ARs in schedule. This work proposes two algorithms 1) Starvation-Removal and 2) AR-to-BE Conversion to solve theseproblems. Experimental results of the proposed algorithms successfully demonstrate that we can stop starvation of BE leasesfor resources and effectively improve request acceptance rate.