Cotes-Ruiz Iván Tomás, Prado Rocío P, García-Galán Sebastián, Muñoz-Expósito José Enrique, Ruiz-Reyes Nicolás
Telecommunication Engineering Department, University of Jaén, Science and Technology Campus, Linares, Spain.
PLoS One. 2017 Jan 13;12(1):e0169803. doi: 10.1371/journal.pone.0169803. eCollection 2017.
Nowadays, the growing computational capabilities of Cloud systems rely on the reduction of the consumed power of their data centers to make them sustainable and economically profitable. The efficient management of computing resources is at the heart of any energy-aware data center and of special relevance is the adaptation of its performance to workload. Intensive computing applications in diverse areas of science generate complex workload called workflows, whose successful management in terms of energy saving is still at its beginning. WorkflowSim is currently one of the most advanced simulators for research on workflows processing, offering advanced features such as task clustering and failure policies. In this work, an expected power-aware extension of WorkflowSim is presented. This new tool integrates a power model based on a computing-plus-communication design to allow the optimization of new management strategies in energy saving considering computing, reconfiguration and networks costs as well as quality of service, and it incorporates the preeminent strategy for on host energy saving: Dynamic Voltage Frequency Scaling (DVFS). The simulator is designed to be consistent in different real scenarios and to include a wide repertory of DVFS governors. Results showing the validity of the simulator in terms of resources utilization, frequency and voltage scaling, power, energy and time saving are presented. Also, results achieved by the intra-host DVFS strategy with different governors are compared to those of the data center using a recent and successful DVFS-based inter-host scheduling strategy as overlapped mechanism to the DVFS intra-host technique.
如今，云系统不断增长的计算能力依赖于降低其数据中心的功耗，以使它们具有可持续性并在经济上盈利。计算资源的高效管理是任何能源感知数据中心的核心，并且特别重要的是使其性能适应工作负载。科学各个领域中的密集计算应用会生成称为工作流的复杂工作负载，而在节能方面对其进行成功管理仍处于起步阶段。WorkflowSim是目前用于工作流处理研究的最先进模拟器之一，具有任务聚类和故障策略等高级功能。在这项工作中，提出了一种预期的WorkflowSim功率感知扩展。这个新工具集成了一个基于计算加通信设计的功率模型，以允许在考虑计算、重新配置和网络成本以及服务质量的情况下优化节能方面的新管理策略，并且它纳入了主机节能的卓越策略：动态电压频率缩放（DVFS）。该模拟器旨在在不同的实际场景中保持一致，并包括广泛的DVFS调速器库。展示了该模拟器在资源利用率、频率和电压缩放、功率、能源和时间节省方面有效性的结果。此外，将使用不同调速器的主机内DVFS策略所取得的结果与使用基于DVFS的最新且成功的主机间调度策略作为主机内DVFS技术的重叠机制的数据中心的结果进行了比较。
