Paradigm-based adaptive provisioning in virtualized data centers

Virtualized data centers host multiple applications with distinct objectives in a shared infrastructure. Accommodating several dynamic applications in virtual data centers is a challenging task for cloud providers. Current provisioning solutions focus on a limited set of objectives that may not be suited for the increasing number of applications deployed in data centers everyday. In this paper we propose an adaptive provisioning architecture for virtualized data centers based on allocation paradigms. A paradigm translates high-level application goals to objectives, allocator instances, and actions that actually provision customized virtual infrastructures to applications. A paradigm policy language is defined to express the relationship between paradigms, objectives, and actions. A performance evaluation of the proposed approach considers four main aspects: acceptance ratio, provisioning cost, and CPU and link utilization. Simulation results show that our proposal is able to select the most appropriate set of allocation actions based on the particularities of the applications.