Section: New Results

Energy efficiency of large scale distributed systems

Participants : Ghislain Landry Tsafack Chetsa, Mohammed El Mehdi Diouri, Jean-Patrick Gelas, Olivier Glück, Laurent Lefèvre, François Rossigneux.

Analysis and Evaluation of Different External and Internal Power Monitoring Devices for a Server and a Desktop Machine

Large-scale distributed systems (e.g., datacenters, HPC systems, clouds, large-scale networks, etc.) consume and will consume enormous amounts of energy. Therefore, accurately monitoring the power and energy consumption of these systems is increasingly more unavoidable. The main novelty of this contribution [15] is the analysis and evaluation of different external and internal power monitoring devices tested using two different computing systems, a server and a desktop machine. Furthermore, we also provide experimental results for a variety of benchmarks which exercise intensively the main components (CPU, Memory, HDDs, and NICs) of the target platforms to validate the accuracy of the equipment in terms of power dispersion and energy consumption. We highlight that external wattmeters do not offer the same measures as internal wattmeters. Thanks to the high sampling rate and to the different measured lines, the internal wattmeters allow an improved visualization of some power fluctuations. However, a high sampling rate is not always necessary to understand the evolution of the power consumption during the execution of a benchmark.

Your Cluster is not Power Homogeneous

Future supercomputers will consume enormous amounts of energy. These very large scale systems will gather many homogeneous clusters. We analyze the power consumption of the nodes from different homogeneous clusters during different workloads. As expected, we observe that these nodes exhibit the same level of performance. However, we also show that different nodes from a homogeneous cluster may exhibit heterogeneous idle power energy consumption even if they are made of identical hardware. Hence, we propose an experimental methodology to understand such differences. We show that CPUs are responsible for such heterogeneity which can reach 20% in terms of energy consumption. So energy aware (Green) schedulers must take care of such hidden heterogeneity in order to propose efficient mapping of tasks. To consume less energy, we propose an energy-aware scheduling approach taking into account the heterogeneous idle power consumption of homogeneous nodes [20] . It shows that we are able to save energy up to 17% while exploiting the high power heterogeneity that may exist in some homogeneous clusters.

Energy Consumption Estimations of Fault Tolerance protocols

Energy consumption and fault tolerance are two interrelated issues to address for designing future exascale systems. Fault tolerance protocols used for checkpointing have different energy consumption depending on parameters like application features, number of processes in the execution and platform characteristics. Currently, the only way to select a protocol for a given execution is to run the application and monitor the energy consumption of different fault tolerance protocols. This is needed for any variation of the execution setting. To avoid this time and energy consuming process, we propose an energy estimation framework [16] , [17] , [7] . It relies on an energy calibration of the considered platform and a user description of the execution setting. We evaluate the accuracy of our estimations with real applications running on a real platform with energy consumption monitoring. Results show that our estimations are highly accurate and allow selecting the best fault tolerant protocol without pre-executing the application.

Energy Consumption Estimations of Data Broadcasting

Future supercomputers will gather hundreds of millions of communicating cores. The movement of data in such systems will be very energy consuming. We address the issue of energy consumption of data broadcasting in such large scale systems. To this end, in [19] , [7] , we propose a framework to estimate the energy consumed by different MPI broadcasting algorithms for various execution settings. Validation results show that our estimations are highly accurate and allow to select the least consuming broadcasting algorithm.

A Smart-Grid Based Framework for Consuming Less and Better in Extreme-Scale Infrastructures

As they will gather hundreds of million cores, future exascale supercomputers will consume enormous amounts of energy. Besides being very important, their power consumption will be dynamic and irregular. Thus, in order to consume energy efficiently, powering such systems will require a permanent negotiation between the energy supplier and one of its major customers represented by exascale platforms. We have designed SESAMES [18] , [53] , a smart and energy-aware service-oriented architecture manager that proposes energy-efficient services for exascale applications and provides an optimized reservation scheduling. The new features of this framework are the design of a smart grid and a multi-criteria green job scheduler. Simulation results show that with the proposed multi-criteria job scheduler, we are able to save up to 2.32 % in terms of energy consumption, 24.22 % in terms of financial cost and reduce up to 7.12 % the emissions of $C{O}_2$.

Clustered Virtual Home Gateway (vHGW)

This result is a joint work between Avalon team (J.P. Gelas, L. Lefevre) and Addis Abeba University (M. Tsibie and T. Assefa). The customer premises equipment (CPE), which provides the interworking functions between the access network and the home network, consumes more than 80% of the total power in a wireline access network. In the GreenTouch initiative (cf Section  8.3 ), we aim at a drastic reduction of the power consumption by means of a passive or quasi-passive CPE. Such approach requires that typical home gateway functions, such as routing, security, and home network management, are moved to a virtual home gateway (vHGW) server in the network. In our first prototype virtual home gateways of the subscribers were put in LXC containers on a unique GNU/Linux server. The container approach is more scalable than separating subscribers by virtual machines. We demonstrated a sharing factor of 500 to 1000 virtual home gateways on one server, which consumes about 150 W, or 150 to 300 mW per subscriber. Comparing this power consumption with the power of about 2 W for the processor in a thick client home gateway, we achieved an efficiency gain of 5-10x. The prototype was integrated and demonstrated at TIA 2012 in Dallas. In our current work, we propose the Clustered vHGWs Data center architecture to yield optimal energy conservation through virtual machine’s migration among physical nodes based on the current subscriber’s service access state, while ensuring SLA respective subscribers. Thus, optimized energy utilization of the data center is assured without compromising the availability of service connectivity and QoS preferences of respective subscribers.

Improving Energy Efficiency of Large Scale Systems without a priori Knowledge of Applications and Services

Unlike their hardware counterpart, software solutions to the energy reduction problem in large scale and distributed infrastructures hardly result in real deployments. At the one hand, this can be justified by the fact that they are application oriented. At the other hand, their failure can be attributed to their complex nature which often requires vast technical knowledge behind proposed solutions and/or thorough understanding of applications at hand. This restricts their use to a limited number of experts, because users usually lack adequate skills. In addition, although subsystems including the memory and the storage are becoming more and more power hungry, current software energy reduction techniques fail to take them into account. We propose a methodology for reducing the energy consumption of large scale and distributed infrastructures. Broken into three steps known as (i) phase identification, (ii) phase characterization, and (iii) phase identification and system reconfiguration; our methodology abstracts away from any individual applications as it focuses on the infrastructure, which it analyses the runtime behaviour and takes reconfiguration decisions accordingly.

The proposed methodology is implemented and evaluated in high performance computing (HPC) clusters of varied sizes through a Multi-Resource Energy Efficient Framework (MREEF). MREEF implements the proposed energy reduction methodology so as to leave users with the choice of implementing their own system reconfiguration decisions depending on their needs. Experimental results show that our methodology reduces the energy consumption of the overall infrastructure of up to 24% with less than 7% performance degradation. By taking into account all subsystems, our experiments demonstrate that the energy reduction problem in large scale and distributed infrastructures can benefit from more than “the traditional” processor frequency scaling. Experiments in clusters of varied sizes demonstrate that MREEF and therefore our methodology can easily be extended to a large number of energy aware clusters. The extension of MREEF to virtualized environments like cloud shows that the proposed methodology goes beyond HPC systems and can be used in many other computing environments.

Reservation based Usage for Energy Efficient Clouds: the Climate Architecture

The FSN XLcloud project (cf Section  8.1 ) strives to establish the demonstration of a High Performance Cloud Computing (HPCC) platform based on OpenStask, that is designed to run a representative set of compute intensive workloads, including more specifically interactive games, interactive simulations and 3D graphics. XLcloud is based on OpenStack, and Avalon is contributing to the energy efficiency part of this project. We have proposed and brought our contribution to Climate, a new resource reservation framework for OpenStack, developed in collaboration with Bull, Mirantis and other OpenStack contributors. Climate allows the reservation of both physical and virtual resources, in order to provide a mono-tenancy environment suitable for HPC applications. Climate chooses the most efficient hosts (flop/W). This metric is computed from the CPU / GPU informations, mixed with real power consumption measurements provided by the Kwapi framework. The user requirements may be loose, allowing Climate to choose the best time slot to place the reservation. Climate will be improved with standby mode features, to shut down automatically the unused hosts. The first release of Climate is planned at the end of January 2014, and we expect an incubation in the next version of OpenStack.