Elastic Resource Allocation in Datacenters: Gremlins in the Management Plane
Virtualization has simplified the management of datacenter infrastructures and enabled new services that can benefit both customers and providers. From a provider perspective, one of the key services in a virtualized datacenter is elastic allocation of resources to work-loads, using a combination of virtual machine migration and per-server work-conserving scheduling. Known challenges to elastic resource allocation include scalability, hardware heterogeneity, hard and soft virtual machine placement constraints, resource partitions, and others. This paper describes an additional challenge, which is the need for IT management to consider two design constraints that are systemic to large-scale deployments: failures in management operations and high variability in cost.