Laurent Lefevre - Academia.edu (original) (raw)

Papers by Laurent Lefevre

Building a sustainable Exascale machine is a very promising target in High Performance Computing ... more Building a sustainable Exascale machine is a very promising target in High Performance Computing (HPC). To tackle the energy consumption challenge while continuing to provide tremendous performance, the HPC community have rapidly adopted GPU-based systems. Today, GPUs have became the most prevailing components in the massively parallel HPC landscape thanks to their high computational power and energy efficiency. Modeling the energy consumption of applications running on GPUs has gained a lot of attention for the last years. Alas, the HPC community lacks simple yet accurate simulators to predict the energy consumption of general purpose GPU applications. In this work, we address the prediction of the energy consumption of CUDA kernels via simulation. We propose in this paper a simple and lightweight energy model that we implemented using the open-source framework SimGrid. Our proposed model is validated across a diverse set of CUDA kernels and on two different NVIDIA GPUs (Tesla M2075 and Kepler K20Xm). As our modeling approach is not based on performance counters or detailed-architecture parameters, we believe that our model can be easily approved by users who take care of the energy consumption of their GPGPU applications.

The heat induced by computing resources is generally a waste of energy in supercomputers. This is... more The heat induced by computing resources is generally a waste of energy in supercomputers. This is especially true in very large scale supercomputers, where the produced heat has to be compensated with expensive and energy consuming cooling systems. Energy is a critical point for future supercomputing trends that currently try to achieve exascale, without having its energy consumption reaching an important fraction of a nuclear power plant. Thus, new ways of generating or recovering energy have to be explored. Energy harvesting consists in recovering wasted energy. ThermoElectric Generators (TEGs) aim to recover energy by converting wasted dissipated energy into usable electricity. By combining computing units (CU) and TEGs at very large scale, we spotted a potential way to recover energy from wasted heat generated by computations on supercomputers. In this paper, we study the potential gains in combining TEGs with computational units at petascale and exascale. We present the technology behind TEGs, the study of a typical supercomputer environment, and finally our results concerning binding TEGs and computational units in a petascale and exascale system. With the available technology, we demonstrate that the use of TEGs in a supercomputer environment could be realistic and quickly profitable, and hence have a positive environmental impact.

The continuous increase of data volumes poses several challenges to established infrastructures i... more The continuous increase of data volumes poses several challenges to established infrastructures in terms of resource management and expenses. One of the most important challenges is the energy-efficient enactment of data operations in the context of data-intensive applications. Computing, generating and exchanging growing volumes of data are costly operations, both in terms of time and energy. In the late literature, different types of compression mechanisms emerge as a new way to reduce time spent on data-related operations, but the overall energy cost has not been studied. Based on current advances and benefits of compression techniques, we propose a model that leverages nonlossy compression and identifies situations where compression presents an interest from an energy reduction perspective. The proposed model considers sender, receiver, communications costs over various types of files and available bandwidth. This strategy allows us to improve both time and energy required for communications by taking advantage of idle times and power states. Evaluation is performed over HPC, Big Data and datacenter scenarios. Results show significant energy savings for all types of file while avoiding counter performances, resulting in a strong incentive to actively leverage non-lossy compression using our model.

OATAO is an open access repository that collects the work of Toulouse researchers and makes it fr... more OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published in : http://oatao.univ-toulouse.fr/ Eprints ID : 17060 The contribution was presented at PRIMA 2016 : Abstract. We present a multi agent system simulating the complex interplay between the actors of innovation involved in the development of technology transfer for Green IT. We focus on the role and the influence of technology transfer offices on the individual objectives of each other actor (researchers, research facilities, companies). We analyse also their impact on several parameters, including sustainability.

OATAO is an open access repository that collects the work of Toulouse researchers and makes it fr... more OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published in : http://oatao.univ-toulouse.fr/ Eprints ID : 18767 The contribution was presented at ICLIE 2016 : Abstract While there is a tremendous increase in academic research and collaboration between academia, the results of exchange between industry and science are steady. To understand this complex situation and to propose an improvement for technology transfer between academia and industry, it is necessary to investigate the different partners involved. We present a multi-agent system to model this technology transfer of green IT in order to see the impact on the development of sustainability in our society. We defi ne a sustainability indicator and we study its changes according to the parameters defi ned in the technology transfer.

While there is a tremendous increase in academic research and collaboration between academia, the... more While there is a tremendous increase in academic research and collaboration between academia, the results of exchange between industry and science are steady. To understand this complex situation and to propose an improvement for technology transfer between academia and industry, it is necessary to investigate the different partners involved. We present a multi-agent system to model this technology transfer of green IT in order to see the impact on the development of sustainability in our society. We defi ne a sustainability indicator and we study its changes according to the parameters defi ned in the technology transfer.

Much of the "big data" generated today is received in near real-time and requires quick... more Much of the "big data" generated today is received in near real-time and requires quick analysis. In Internet of Things (IoT) [1, 9], for instance, continuous data streams produced by multiple sources must be handled under very short delays. As a result, several stream processing engines have been proposed. Under several engines, a stream processing application is a directed graph or dataflow whose vertices are operators that execute a function over the incoming data and edges that define how data flows between them. A dataflow has one or multiple sources (i.e., sensors, gateways or actuators), operators that perform transformations on the data (e.g., filtering, mapping, and aggregation) and sinks (i.e., queries that consume or store the data). In a traditional cloud deployment, the whole application is placed in the cloud computing to benefit from virtually unlimited resources. However, processing all the data in the cloud can introduce latency due to data transfer, which...

2017 13th International Conference on Network and Service Management (CNSM), 2017

Lecture Notes in Computer Science, 2016

We present a multi agent system simulating the complex interplay between the actors of innovation... more We present a multi agent system simulating the complex interplay between the actors of innovation involved in the development of technology transfer for Green IT. We focus on the role and the influence of technology transfer offices on the individual objectives of each other actor (researchers, research facilities, companies). We analyse also their impact on several parameters, including sustainability.

2008 Ninth International Conference on Parallel and Distributed Computing, Applications and Technologies, 2008

Fault tolerance can be defined as the capability of a system or a component to continue normal op... more Fault tolerance can be defined as the capability of a system or a component to continue normal operation, despite the occurrence of a hardware or a software fault. Frameworks providing a fault tolerant service take advantage of resource redundancy to provide high availability capabilities. One of the critical challenges that led to this work is the observation that the existing fault tolerance frameworks are not adapted to current and next generation Internet services. Indeed, they do not provide consistent serviceaware failure recovery capabilities. Particularly, little interest has been granted to transport level awareness despite its important partaking in improving the reliability of connection oriented services and stateful devices. In this paper, we evaluate an active replication based framework for highly available Internet services. The proposed framework is fully client/server transparent. Performance evaluations show that it incurs minimal overhead to end-to-end conversations during failsafe periods and performs well during failures.

This paper presents the design, implementation, and evaluation of a new technique aimed to reduce... more This paper presents the design, implementation, and evaluation of a new technique aimed to reduce the number of interrupts due to transmitted packets in the parallel network subsystem KNET. This technique combines interrupt and polling modes and relies on support in the Network Interface Controller-per-processor transmit queues-for distributing processing of transmission notifications among processors. Our prototype implementation on Linux and Myrinet yields performance gains of 16% on a 4-processor machine running a web server application.

This document describes a new networking subsystem architecture built around a packet classifier ... more This document describes a new networking subsystem architecture built around a packet classifier executing in the Network Interface Card (NIC). By classifying packets in the NIC, we believe that performance, scalability, and robustness can be significantly improved on shared-memory multiprocessor Internet servers. In order to demonstrate the feasibility and the benefits of the approach, we developed a software prototype (consisting in extensions to the Linux kernel and modifications to the Myrinet NIC firmware and driver) and ran a series of experiments. The obtained results, presented therein, show the relevance of the approach.

2008 5th IEEE Consumer Communications and Networking Conference, 2008

Journal of Grid Computing, 2010

Grids organize resource sharing, a fundamental requirement of large scientific collaborations. Se... more Grids organize resource sharing, a fundamental requirement of large scientific collaborations. Seamless integration of grids into everyday use requires responsiveness, which can be provided by elastic Clouds, in the Infrastructure as a Service (IaaS) paradigm. This paper proposes a model-free resource provisioning strategy supporting both requirements. Provisioning is modeled as a continuous action-state space, multi-objective reinforcement learning (RL) problem, under realistic hypotheses; simple utility functions capture the high level goals of users, administrators, and shareholders. The model-free approach falls under the general program of autonomic computing, where the incremental learning of the value function associated with the RL model provides the so-called feedback loop. The RL model includes an approximation of the value function through an Echo State Network. Experimental validation on a real dataset from the EGEE grid shows that introducing a moderate level of elasticity is critical to ensure a high level of user satisfaction.

21st International Conference on Advanced Networking and Applications (AINA '07), 2007

HAL (Le Centre pour la Communication Scientifique Directe), 2019

Face à la croissance massive des échanges de données et des besoins de stockage, l'impact spatial... more Face à la croissance massive des échanges de données et des besoins de stockage, l'impact spatial et énergétique des data centers va être de plus en plus structurant pour les territoires. Leur diversité d'usages, d'acteurs, de tailles et d'implantations rend aujourd'hui complexe la lecture de leurs dynamiques et de leurs effets spatiaux. Le présent rapport s'attache donc à donner une image du paysage des data centers en Ile-de-France et dans trois territoires des États-Unis, représentant chacun des situations spatiales et énergétiques différentes (ville dense, espace périphérique, rural). Facteur potentiel de déséquilibre des systèmes énergétiques locaux, objets dont l'accumulation urbaine et la dispersion rurale questionnent, les data centers font ici l'objet d'une analyse approfondie pour mieux appréhender les nouveaux territoires numériques en construction, les solidarités énergétiques à construire et les alliances d'acteurs à mettre en place. Un focus est également réalisé sur les infrastructures numériques alternatives et citoyennes, qui se développent aussi bien en Afrique, Amérique du Sud, que dans les territoires mal couverts en Europe ou aux États-Unis. Dédiées à l'accès à Internet et de plus en plus, aux services d'hébergement et de cloud, elles peuvent constituer une réponse distribuée et pair-à-pair, dont l'impact écologique pourrait finalement se révéler plus limité que les infrastructures centralisées de grande échelle car calibrées au plus près des besoins locaux, mais aussi plus résilientes car moins centralisées techniquement et moins concentrées spatialement. Elles constituent ainsi une option à considérer, soutenir mais aussi à mieux évaluer, pour réduire les impacts spatiaux et énergétiques des data centers. Le rapport propose également des visions prospectives qui combinent des tendances de fond et des signaux faibles pour imaginer les mondes numériques de demain, dont trois possibles sont décrits : « Croissance et ultracentralisation numérique » ; « Stabilisation du Système Technique Numérique et diversité infrastructurelle : quête d'une difficile résilience » ; « Ultradécentralisation numérique : la fin des data centers ? ». Enfin, des recommandations sont proposées autour de 3 axes : les acteurs et la gouvernance ; l'urbanisme et l'environnement ; l'énergie. Des pistes d'approfondissement et d'études sont également présentées.

2018 11th International Conference on the Quality of Information and Communications Technology (QUATIC), 2018

The quality of services in Cloud Computing (CC) depends on the scheduling strategies selected for... more The quality of services in Cloud Computing (CC) depends on the scheduling strategies selected for processing of the complex workloads in the physical cloud clusters. Using the scheduler of the single type does not guarantee of the optimal mapping of jobs onto cloud resources, especially in the case of the processing of the big data workloads. In this paper, we compare the performances of the cloud schedulers for various combinations of the cloud workloads with different characteristics. We define several scenarios where the proper types of schedulers can be selected from a list of scheduling models implemented in the system, and used to schedule the concrete workloads based on the workloads' parameters and the feedback on the efficiency of the schedulers. The presented work is the first step in the development and implementation of an automatic intelligent scheduler selection system. In our simple experimental analysis, we confirm the usefulness of such a system in today's data-intensive cloud computing.

Future Generation Computer Systems, 2022

This paper focuses on SDN-based approaches for deploying stream processing workloads on heterogen... more This paper focuses on SDN-based approaches for deploying stream processing workloads on heterogeneous environments comprising wide-area networks, cloud and fog resources. Stream processing applications impose strict latency requirements to operate appropriately. Deploying workloads in the fog reduces unnecessary delays, but its computational resources may not handle all the tasks. On the other hand, offloading the tasks to the cloud is constrained by limited network resources and involves additional transmission delays that exceed latency thresholds. Adaptive workload deployment may solve these issues by ensuring that resource and latency requirements are satisfied for all the data streams processed by an application. This paper's main contribution consists of dynamic workload placement algorithms operating on stream processing requests with latency constraints. Provisioning of computing infrastructure exploits the interplay between fog and cloud under limited network capacity. The algorithms aim to maximize the ratio of successfully handled requests by effectively utilizing available resources while meeting application latency constraints. Experiments demonstrate that the goal can be achieved by detailed analysis of requests and ensuring balanced computing and network resources utilization. As a result, up to 30% improvement over the reference algorithms in success rate is observed.

This short paper presents how to integrate the Disruption

OATAO is an open access repository that collects the work of Toulouse researchers and makes it fr... more OATAO is an open access repository that collects the work of Toulouse researchers and makes it freely available over the web where possible. This is an author-deposited version published in : http://oatao.univ-toulouse.fr/ Eprints ID : 18767 The contribution was presented at ICLIE 2016 : Abstract While there is a tremendous increase in academic research and collaboration between academia, the results of exchange between industry and science are steady. To understand this complex situation and to propose an improvement for technology transfer between academia and industry, it is necessary to investigate the different partners involved. We present a multi-agent system to model this technology transfer of green IT in order to see the impact on the development of sustainability in our society. We defi ne a sustainability indicator and we study its changes according to the parameters defi ned in the technology transfer.