Section: New Results

Energy-efficient Resource Infrastructures

Participants : Alexandra Carpen-Amarie, Bogdan Florin Cornea, Ismael Cuadrado Cordero, Djawida Dib, Eugen Feller, Yunbo Li, Christine Morin, Anne-Cécile Orgerie, Guillaume Pierre.

Energy-efficient IaaS clouds

Participants : Alexandra Carpen-Amarie, Christine Morin, Anne-Cécile Orgerie.

Energy consumption has always been a major concern in the design and cost of data centers. The wide adoption of virtualization and cloud computing has added another layer of complexity to enabling an energy-efficient use of computing power in large-scale settings. Among the many aspects that influence the energy consumption of a cloud system, the hardware-component level is one of the most intensively studied. However, higher-level factors such as virtual machine properties, their placement policies or application workloads may play an essential role in defining the power consumption profile of a given cloud system. In this work, we explored the energy consumption patterns of Infrastructure-as-a-Service (IaaS) cloud environments under various synthetic and real application workloads. For each scenario, we investigated the power overhead triggered by different types of virtual machines, the impact of the virtual cluster size on the energy-efficiency of the hosting infrastructure and the tradeoff between performance and energy consumption of MapReduce virtual clusters through typical cloud applications [21] .

Energy-aware IaaS-PaaS co-design

Participants : Alexandra Carpen-Amarie, Djawida Dib, Guillaume Pierre, Anne-Cécile Orgerie.

The wide adoption of the cloud computing paradigm plays a crucial role in the ever-increasing demand for energy-efficient data centers. Driven by this requirement, cloud providers resort to a variety of techniques to improve energy usage at each level of the cloud computing stack. However, prior studies mostly consider resource-level energy optimizations in IaaS clouds, overlooking the workload-related information locked at higher levels, such as PaaS clouds. We argue that cross-layer cooperation in clouds is a key to achieving an optimized resource management, both performance and energy-wise. To this end, we claim there is a need for a cooperation API between IaaS and PaaS clouds, enabling each layer to share specific information and to trigger correlated decisions. We identified the drawbacks raised by such co-design objectives and discuss opportunities for energy usage optimizations, and plan to start the research to address these issues in 2014.

Performance and energy-efficiency evaluation of Hadoop deployment models

Participants : Eugen Feller, Christine Morin.

The exponential growth of scientific and business data has resulted in the evolution of the cloud computing and the MapReduce parallel programming model. Cloud computing emphasizes increased utilization and power savings through consolidation while MapReduce enables large scale data analysis. The Hadoop framework is the most popular open source software implementing the MapReduce model. In our work, we evaluated Hadoop performance in two modes – the traditional model of collocated data and compute services and separated mode where the services are deployed on separate services. The separation of data and compute services provides more flexibility in environments where data locality might not have a considerable impact such as virtualized environments and clusters with advanced networks. In this work, we also conducted an energy efficiency evaluation of Hadoop on physical and virtual clusters in different configurations. The experiments were performed on the Grid’5000 experimentation testbed. To enable virtual machine management, the Snooze cloud stack developed by the Myriads project-team was used. Our extensive evaluation shows that: (1) performance on physical clusters is significantly better than on virtual clusters; (2) performance degradation due to separation of the services depends on the data to compute ratio; (3) application completion progress correlates with the power consumption and power consumption is heavily application specific  [28] .

Energy consumption models and predictions for large-scale systems

Participant : Christine Morin.

Responsible, efficient and well-planned power consumption is becoming a necessity for monetary returns and scalability of computing infrastructures. While there is a variety of sources from which power data can be obtained, analyzing this data is an intrinsically hard task. In our work, we described a generic approach to analyze large power consumption datasets collected from computing infrastructures. As a first step, we proposed a data analysis pipeline that can handle the large-scale collection of energy consumption logs, apply sophisticated modeling to enable accurate prediction, and evaluate the efficiency of the analysis approach. We presented the analysis of a power consumption data set collected over a 6-month period from two clusters of the Grid’5000 experimentation platform used in production. We used Hadoop with Pig to handle the large volume of data. Our data processing generated a summary of the data that provides basic statistical aggregations, over different time scales. The aggregate data was then analyzed as a time series using sophisticated modeling methods with R statistical software. We exploited time series to detect outliers and derive hourly and daily power consumption predictive models. We demonstrated the accuracy of the predictive models and the efficiency of the data processing performed on a 55-node cluster at NERSC [34] . Energy models from such large dataset can help in understanding the evolution of consumption patterns, predicting future energy trends, and providing basis for generalizing the energy models to similar large-scale systems.

Simulating Energy Consumption of Wired Networks

Participant : Anne-Cécile Orgerie.

Predicting the performance of applications, in terms of completion time and resource usage for instance, is critical to appropriately dimension resources that will be allocated to these applications. Current applications, such as web servers and Cloud services, require lots of computing and networking resources. Yet, these resource demands are highly fluctuating over time. Thus, adequately and dynamically dimension these resources is challenging and crucial to guarantee performance and cost-effectiveness. In the same manner, estimating the energy consumption of applications deployed over heterogeneous cloud resources is important in order to provision power resources and make use of renewable energies. Concerning the consumption of entire infrastructures, some studies show that computing resources represent the biggest part in Cloud’s consumption, while others show that, depending on the studied scenario, the energy cost of the network infrastructure that links the user to the computing resources can be bigger than the energy cost of the servers. In this work, we aim at simulating the energy consumption of wired networks which receive little attention in the Cloud computing community even though they represent key elements of these distributed architectures. To this end, we are contributing to the well-known open-source simulator ns3 by developing an energy consumption module named ECOFEN.

Simulating the impact of DVFS within SimGrid

Participants : Alexandra Carpen-Amarie, Christine Morin, Anne-Cécile Orgerie.

Simulation is a a popular approach for studying the performance of HPC applications in a variety of scenarios. However, simulators do not typically provide insights on the energy consumption of the simulated platforms. Furthermore, studying the impact of application configuration choices on energy is a difficult task, as not many platforms are equipped with the proper power measurement tools. The goal of this work is to enable energy-aware experimentations within the SimGrid simulation toolkit, by introducing a model of application energy consumption and enabling the use of DVFS techniques for the simulated platforms. We provide the methodology used to obtain accurate energy estimations, highlighting the simulator calibration phase. The proposed energy model is validated by means of a large set of experiments featuring several benchmarks and scientific applications. This work is available in the latest SimGrid release.