Time-Sensitive Virtual Machines Provisioning and Resource Allocation in Clouds

To efficiently utilize the resources in cloud computing systems, many allocation algorithms have been studied. In this work, for user requests that have timing constraints (i.e., deadlines), we study time-sensitive resource allocation and virtual machine provisioning schemes in cloud systems. Specifically, considering a cloud system with multiple pools of resources (e.g., computing cores and memory) where the resources are provisioned to user requests in the form of virtual machines (VMs), we first study the earliest deadline first (EDF) resource allocation scheme where the request with the earliest deadline is served first. Then, by incorporating both resource usage efficiency and deadlines of requests, we propose a time-sensitive resource (TSR) allocation scheme. A unified scheme that integrates the ideas of both EDF and TSR is also studied. By incorporating the classical mapping schemes (i.e., first-fit, best-fit and worst fit) when selecting the pool of resources, the proposed schemes are evaluated through extensive simulations using real-application trace data. The results show that the proposed schemes can significantly out-perform the state-of-the-art deadline oblivious resource allocation scheme with up to 25% more user requests being served and up to 5% more benefit being achieved, especially for over-loaded cloud systems.