Merkat: Market-based Autonomous Application and Resource Management in the Cloud Merkat: Market-based Autonomous Application and Resource Management in the Cloud

Organizations owning High Performance Computing (HPC) infrastructures are facing difficulties in managing their resources. These difficulties come from the need to provide concurrent resource access to applications with different resource requirements while considering that users are selfish and might have different performance objectives, or Service Level Objectives (SLOs), when executing them. To address these challenges, this paper proposes Merkat, a market-based SLO-driven cloud platform. Merkat relies on a market-based model to allocate resources to ap- plications while taking advantage of virtualization technologies and on-demand provisioning to maximize resource utilization. Merkat’s resource market uses a combination of currency distribu- tion and dynamic resource pricing to ensure proper resource distribution while decentralizing the resource control. In Merkat autonomous controllers apply adaptation policies to scale the appli- cation’s resource demand according to user’s SLO. The adaptation policies can: (i) dynamically tune the amount of CPU and memory provisioned for the virtual machines in contention periods; (ii) dynamically change the number of virtual machines. We evaluated this proposed platform in simulation and on the Grid’5000 testbed. Results show that: (i) Merkat provides flexible suport for different application types and different SLOs; (ii) it increases the resource utilization of the infras- tructure; (iii) and is capable of providing good user satisfaction compared to existing centralized systems.