Research On Energy-efficient Resource Allocation In Cloud Data Centers

Cloud Computing a new computing paradigm, which connects all kinds of re-sources through a communication network and provides a shared resource pool in on-demand self-service manner. Data center is the physical entity of cloud comput-ing, and is one of the key factors that promote the development of cloud computing. As the development of cloud computing, the energy consumption issue of cloud data center has attracted extreme concern from the communities. Resource allocation, which affects the resource utilization mode and task execution mode, has a signif-icant effect on the energy consumption of physical machines (PMs), network, and task execution, which account most of total cloud data center consumption.In this paper, we aim to achieve energy conservation from the aspect of re-source utilization and task execution, and propose resource allocation mechanism to optimize energy consumption on PMs, network, and task execution, respectively. The main contributions are as follows:1. We present a balanced utilization of multi-dimensional resource based resource allocation mechanism for the online VM placement problem. We propose a CPU-utilization based energy consumption model for PM. We find that there may be resource leak for the PM, while dealing with online VM requests, and present a multi-dimensional resource utilization model to characterize the resource usage quantitatively. Based on this model, we present a VM place-ment algorithm that can increase the resource utilization. The simulations and experiments on real data traces show that the multi-dimensional resource utilization based VM placement algorithm can lead to reasonable cloud data center resource allocation, and has a significant effect on energy saving for the running of PMs.2. We present efficient network-aware resource allocation mechanism for the com-plex VM placement problem. A series of VM placement algorithms based on recursion are proposed, and we analyze the cloud data center energy consump-tion optimization under various communication models theoretically. We first study the VM placement problem to minimize the network energy consump-tion for given PMs, and further classify the problem into homogeneous case and heterogeneous case. For the homogeneous case, we present optimal algorithms under various communication models. We conduct theoretic performance anal- ysis and prove the optimality of the algorithms. For the heterogeneous case, we present an approximation algorithm, and analyze the approximation ratio un-der various communication models. For the case that take into account energy consumption for both PMs and network, we study how the energy changes as various numbers of PMs are used, and present an efficient binary search based heuristic algorithm to achieve the optimal number of PMs, and conduct the VM placement based on the determined PMs. We conduct extensive simu-lations to evaluate the performance of algorithms, and they have significant energy saving for the cloud data center.3. We present optimized computational and storage resource allocation mechanis-m for the jointly data-intensive task and data placement problem. We evaluate the objectivity and significance of data locality by extensive real experiments. Hence, we formulate the jointly placement problem while preserving full da-ta locality for all tasks, which can reduce the task execution time and avoid data transferring energy consumption. It is an efficient way to achieve energy conservation for task execution. We find that there is an indicator to char-acterize the hardness of the problem quantitatively. The indicator is based on the number of data blocks and the number of servers with fixed memory slots, and has no relationship with the number of jobs. According to the in-dicator, we classify the problem into various cases, and analyze their hardness respectively. For the case that proved to be NP-hard, we present a heuristic algorithm and a 2-approximation algorithm. For most of the cases, we prove that the placement problem can be solved optimally, and present an optimal algorithm that lead to least task execution energy consumption.