Joint Optimization Of Resource Allocation And Scheduling For Cloud-based Video Streaming Service Systems

With the rapid development of high-speed Internet and media technologies, online video services have become one of the most popular Internet applications. Traditionally, video service providers build their large-scale video streaming service systems mainly based on the content distribution network (CDN) and peer to peer network (P2P) ar-chitectures. However, the CDN’s semi-static resource allocation mechanism makes the video service system have poor scalability and low resource utilization rate, which will incur a high service cost. These disadvantages make it difficult for CDN to cope with the increasing number of users requests. P2P relies on users to cooperatively upload data, which is difficult to ensure the reliability of service, so P2P can not guarantee the requirements of users for higher video viewing experience. Compared with tradition-al technologies, the emerging cloud computing technology provides a reliable, flexible and cost-effective resource allocation method, which brings a new solution for video service providers. In this paper, we investigate to build a new generation of large-scale video streaming service systems over cloud computing platforms. The video service providers dynamically adjust the allocated cloud computing resources in an on-demand manner to cope with highly dynamic and heterogeneous video requests. The elastic resource provisioning can reduce the operating costs and improve resource utilization.In this paper we investigate several important problems for the cloud-based video service system, including dynamic resource allocation, request scheduling, content place-ment and network management. We establish a mathematical model to describe the system and utilize several related optimization theories to derive the optimal control strategy. The main contributions of this paper are summarized as follows:1)We consider that a cloud computing service provider provides various computing resources to video service providers by leasing virtual machines, and offers multiple pricing models for virtual machines. In order to reduce operating costs, video service providers need to dynamically adjust the number of the rented virtual machines to serve the users’request. In this paper, the above problem is modeled as solving the minimum number of virtual machines with quality of service(QoS) constraints and the optimal procurement strategy under multiple pricing models.The QoS constraint is expressed as that the overload probability of the system should be bounded by a preset threshold. We use the large derivation theory to derive an estimate model for the overload probability and obtain the optimal allocation of the virtual machine based on the online measurement. Then, by comparing various pricing models, we solve the integer programming to find out the optimal procurement scheme of virtual machines for each time slot. Further, we propose a long-term dynamic adjustment strategy for the reserved virtual machines. Finally, we conduct extensive simulations to validate the effectiveness of our algorithms in the realistic settings.2) We investigate a large-scale video service system deployed on multiple geograph-ically distributed cloud datacenters which delivers various videos to users spread over multiple regions.In order to improve users’quality of experience (QoE) and reduce the operational cost, the video service provider needs to constantly adjust the video placement and resource allocation while dispatching user requests to appro-priate datacenters according to a certain criterion. We model the system operational process as a Markov decision process and use the average performance criterion to represent the average profit over a long run of the system. We use a utility func-tion to transform the users’QoE into the system revenue and the profit is equal to the revenue minus the cost. Then we formulate the optimization problem as finding the joint optimal request dispatching and video placement policy which maximizes the average profit. Based on the performance sensitivity analysis, we propose a sample-path-based policy iteration algorithm to obtain the optimal policy and prove its optimality. Then we discuss the practical implementation of our algorithm and conduct simulations to evaluate the performance of our algorithm.3) Cloud computing along with Software-Defined Networking (SDN) provides an op-portunity of integration of resource allocation and network management. We in-troduce an SDN-enabled cloud mobile video distribution architecture and the data-centers are interconnected by the SDN-enabled network. We propose a joint video placement, request dispatching and traffic management mechanism to improve us-er experience and reduce the system operational cost. We use a utility function to capture the two aspects of user experience:the level of satisfaction and average la-tency. We integrate request dispatching and response path routing as a unified model of traffic management and formulate the joint optimization problem as a mixed inte-ger programming (MIP). Based on the dual decomposition and subgradient methods, we develop an optimal algorithm to solve the MIP problem and prove its optimality. We conduct simulations to evaluate the performance of our algorithm and the results validate the effectiveness of our optimal strategy.