Cost, energy, and response delay awareness-solution for cloud resources management: proposition of a predictive dynamic algorithm for VMs allocation over a distributed cloud infrastructure

Virtual resources allocation and placement problem in a distributed Cloud infrastructure presents a compromising question. The geographic position of data centers; their available free resources; the correspondent delay and energy consumption constraints are factors that involve determining the best allocation and placement decision. Allocation and placement cost will be relatively determined according to that choice. In fact, data centers, installed in cold regions, offer lower costs because they need few cooling maintenances, consequently, energy consumptions are minimized. However, data centers, installed closer to population areas, could impose higher costs because of their limited resources or high need for cooling maintenance, so energy consumptions have to be higher. On the other hand, and within acceptable network conditions, allocating powerful resources placed in closer data centers may guarantee shorter global response delay. This could be helpful to support delay-sensitive applications such as Massively Multi-players Online Gaming (MMOG) and enhance their relative Quality of Experience (QoE). However, it may engender high costs and vice versa. In this view, the present paper highlights the critical relationship between the three basics metrics affecting the QoE of the MMOG service, namely the cost, the energy consumption, and the global response delay. We propose a Predictive Dynamic Virtual Machines (VMs) Allocation and Placement algorithm based on the Seasonal Autoregressive Integrated Moving Average (SARIMA) prediction model that captures the intrinsic trade-off of these metrics and outcomes the best mapping of necessary allocated resources. Our contribution is formulated as a Multiple Multidimensional Knapsack Problem (MMKP). Results show the effectiveness of our contribution in maintaining the balance between low-cost objective, low energy consumption by minimizing the inter-migrations of VMs over data centers, and acceptable delay maintained under a predefined threshold.