Energy Efﬁcient Resource Allocation for Virtual Network Services with Dynamic Workload in Cloud Data Centers
With the rapid proliferation of cloud computing, more and more network services and applications are deployed on cloud data centers. Their energy consumption and green house gas emissions have signiﬁcantly increased. Some efforts have been made to control and lower energy consumption of data centers such as, proportional energy consuming hardware, dynamic provisioning, and virtualization machine techniques. However, it is still common that many servers and network resources are often underutilized, and idle servers spend a large portion of their peak power consumption. Network virtualization and resource sharing have been employed to improve energy efﬁciency of data centers by aggregating workload to a few physical nodes and switch the idle nodes to sleep mode. Especially, with the advent of live migration, a virtual node can be moved from one physical node to another physical node without service disrup tion. It is possible to save more energy by shrinking virtual nodes to a small set of physical nodes and turning the idle nodes to sleep mode when the service workload is low, and expanding virtual nodes to a large set of physical nodes to satisfy QoS requirements when the service workload is high. When the service provider explicates the desired virtual network including a speciﬁc topology, and a set of virtual nodes with certain resource demands, the infrastructure provider computes how the given virtual network is embedded to its operated data centers with minimum energy consumption. When the service provider only gives some description about the network service and the desired QoS requirements, the infrastructure provider has more freedom on how to allocate resources for the network service. For the ﬁrst problem, we consider the evolving workload of the virtual networks or virtual applications and residual resources in data centers, and build a novel model of energy efﬁcient virtual network embedding (EE-VNE) in order to minimize energy usage in the physical network consists of multiple data centers. In this model, both operation cost for executing network services’ task and migration cost for the live migrations of virtual nodes are counted toward the total energy consumption. In addition, rather than random generated physical network topology, we use practical assumption about physical network topology in our model. Due to the NP-hardness of the proposed model, we develop a heuristic algorithm for virtual network scheduling and mapping. In doing so, we speciﬁcally take the expected energy consumption at different times, virtual network operation and future migration costs, and a data center architecture into consideration. Our extensive evaluation results showthatouralgorithmcouldreduceenergyconsumptionupto40%andtakeuptoa57% higher number of virtual network requests over other existing virtual mapping schemes. However, through comparison with CPLEX based exact algorithm, we identify that there is still a gap between the heuristic solution and the optimal solution. Therefore, after investigation other solutions, we convert the origin EE-VNE problem to an Ant Colony Optimization (ACO) problem by building the construction model and presenting the transition probability formula. Then, ACO based algorithm has been adapted to solve the ACO-EE-VNE problem. In addition, we reduce the space complexity of ACO-EE VNE by developing a novel way to track and update the pheromone. For the second problem, we design a framework to dynamically allocate resources for a network service by employing container based virtual nodes. In the framework,each network service would have a pallet container and a set of execution containers. The pal let container requests resource based on certain strategy, creates execution containers with assigned resources and manage the life cycle of the containers; while the execution containers execute the assigned job for the network service. Formulations are presented to optimize resource usage efﬁciency and save energy consumption for network services with dynamic workload, and a heuristic algorithm is proposed to solve the optimization problem. Our numerical results show that container based resource allocation provide more ﬂexible and saves more cost than virtual service deployment with ﬁxed virtual machines and demands. In addition, we study the content distribution problem with joint optimization goal and varied size of contents in cloud storage. Previous research on content distribution mainly focuses on reducing latency experienced by content customers. A few recent studies address the issue of bandwidth usage in CDNs, as the bandwidth consumption is an important issue due to its relevance to the cost of content providers. However, few researches consider both bandwidth consumption and delay performance for the content providers that use cloud storages with limited budgets, which is the focus of this study. We develop an efﬁcient light-weight approximation algorithm toward the joint optimization problem of content placement. We also conduct the analysis of its theoretical complexities. The performance bound of the proposed approximation algorithm exhibits a much better worst case than those in previous studies. We further extend the approximate algorithm into a distributed version that allows it to promptly react to dynamic changes in users’ interests. The extensive results from both simulations and Planetlab experiments exhibit that the performance is near optimal for most of the practical conditions.
Table of Contents
Introduction -- Related work -- Energy efficient virtual network embedding for green data centers using data center topology and future migration -- Ant colony optimization based energy efficient virtual network embedding -- Energy aware container based resource allocation for virtual services in green data centers -- Achieving optimal content delivery using cloud storage -- Conclusions and future work