Dynamic optimal resource provisioning for VoD services under Amazon EC2's pricing models

For video on demand (VoD) services, traditional CDN-based (Content Delivery Network) architecture is not an effective solution to meet high dynamic user demands. Compared with the CDN's semi-static resource allocation, cloud computing provides an elastic and effective resource provisioning mechanism that video service providers can scale up and down the resources rented on demand from cloud data centers to handle the fluctuations of user demands. Thus, it is a challenge for a video service provider to determine the optimal number of virtual machines under multiple pricing model while maintaining quality of service (QoS). In this paper, we propose a cloud-based VoD system model and investigate the resource provisioning problem under the QoS constraint which is defined as overload probability. By applying the large deviation principle, we derive an overload probability estimation model and design an online VM provisioning strategy to make near-optimal VM instances procurement plan at every time slot. We conduct simulations to evaluate our algorithm using real video watching statistics data and the results show that our strategy can effectively cut down the total cost and guarantee QoS.