Defragmentation of Resources in Virtual Desktop Clouds for Cost-Aware Utility-Optimal Allocation

Cloud Service Providers (CSPs) make virtual desktop cloud (VDC) resource provisioning decisions within desktop pools based on user groups and their application profiles. Such provisioning is aimed to satisfy acceptable user quality of experience (QoE) levels and is coupled with subsequent placement of VDs across distributed data centers. The placement decisions are influenced by session latency, load balancing and operation cost constraints. In this paper, we identify the resource fragmentation problem that occurs when placement is done opportunistically to minimize provisioning time and deliver satisfactory user QoE. To solve this problem, which inherently is an NP-Hard problem, we propose a defragmentation scheme that has fast convergence time and has three levels of complexity: (i) "utility fair provisioning" (UFP) to optimize resource provisioning within a data center - to achieve relative fairness between desktop pools, (ii) "static migration-free utility optimal placement and provisioning" (MUPP) to optimize resource provisioning between multiple data centers - to improve performance, and (iii) "dynamic global utility optimal placement and provisioning" (GUPP) to optimize resource provisioning using cost-aware and utility-maximal VD re-allocations and migrations - to increase scalability. We evaluate our defragmentation scheme against \´least latency\´, \´least load\´, and \´least cost\´ schemes using a novel "VDC-Sim" simulator that we have developed in this study. Our simulations leverage profiles of user groups and their applications within desktop pools, obtained from a real VDC test bed. Our simulation results demonstrate that defragmentation is an important optimization step that can enable CSPs to achieve fairness, substantially improve user QoE and increase VDC scalability.