Kingfisher: A System for Elastic Cost-aware Provisioning in the Cloud

Cloud computing platforms allow application providers to rent server capacity to run hosted applications and to dynamically vary the rented capacity to match demand. Today’s cloud platforms offer a plethora of different server configurations for rent and price them differently on a cost-per-core basis. Furthermore, cloud platforms support different replication and migration mechanisms to support elastic provisioning of servers. In this paper, we present Kingfisher a cost-aware provisioning system for cloud applications that can optimize either the rental cost for provisioning a certain capacity or the transition cost of reconfiguring an application’s current capacity. Our system exploits both replication and migration to dynamically provision capacity and uses an integer linear program formulation to optimize cost. We have implemented a prototype of our Kingfisher cloud provisioning system and have evaluated its efficacy on a laboratorybased private Xen cloud as well on the public Amazon EC2 cloud. Our experiments demonstrate the efficacy of Kingfisher in elastically provisioning servers within private and public clouds that see varying application workloads. Our results demonstrate the ability of Kingfisher in reducing server rental costs and reconfiguration overheads over prior costoblivious approaches.