AAGA: Affinity-Aware Grouping for Allocation of Virtual Machines

Virtualization technology enables various application services to be distributed and encapsulated within virtual machines (VMs), which are dynamically allocated to physical machines (PMs) in cloud computing environments. However, in many existing virtualized systems, the limited network bandwidth often becomes a bottleneck resource, leading to the intensification of network competition and the performance degradation for communication or data intensive applications. Aiming at reducing communication overheads and improving the application performance, in this paper, we propose an Affinity-Aware Grouping method for Allocation of VMs (AAGA). Firstly, we identity and model the problem of affinity-aware grouping-based allocation for virtual machines, and propose a detailed grouping method based on which a heuristic bin packing algorithm is used to deploy VM groups into PMs. In order to demonstrate the effectiveness of AAGA, we create multiple real virtual clusters (multi-VCs) with 56 VMs running multi-VM applications and compare application performance with Non-Affinity-aware Grouping-based Allocation methods (NAGA). Experimental results show that AAGA achieves better performance than NAGA.