Calendar

Events in June–July 2018

  • Keynote du LIG: Erol Gelenbe

    Category: Seminars Keynote du LIG: Erol Gelenbe


    June 7, 2018

    TBA

  • Towards a control-theory approach for minimizing unused grid resources, by Agustín Yabo (Master 2, Datamove + Ctrl-A)

    Category: Seminars Towards a control-theory approach for minimizing unused grid resources, by Agustín Yabo (Master 2, Datamove + Ctrl-A)


    June 14, 2018

    HPC systems are facing more and more variability in their behavior, related to e.g., performance and power consumption, and the fact that they are less predictable requires more runtime management. This can be done in an Autonomic Management feedback loop, in response to monitored information in the systems, by analysis of this data and utilization of the results in order to activate appropriate system-level or application-level feedback mechanisms (e.g., informing schedulers, down-clocking CPUs).
    Such problem is found in the context of CiGri, a simple, lightweight, scalable and fault tolerant grid system which exploits the unused resources of a set of computing clusters. Computing power left over by the execution of a main HPC application scheduling is used to execute smaller jobs, which are injected as much as the global system allows.
    The seminar will presents results addressing the problem of automated resource management in an HPC infrastructure, using techniques from Control Theory to design a controller that maximizes cluster utilization while considering cluster and fileserver overloading. We draw from a previous work where a proportional-integral feedback (Proportional Integral, PI) control loop was implemented, through a maximum number of jobs to be sent to the cluster, in response to system information about the current number of jobs processed. Then, we proposed a dynamical model for the system queue, cluster and fileserver load with time-varying parameters, an EKF (Extended Kalman Filter) for performing online parameter estimation and an MPC (Model-Predictive Control) approach for regulating the desired values of the infrastracture to the desired ones.

    Bat. IMAG, 206
  • Pré-soutenances M2R DATAMOVE/POLARIS

    Category: Seminars Pré-soutenances M2R DATAMOVE/POLARIS


    June 21, 2018

    1. Emelianov Vitalli
    2. Janon Alexis
    3. Colao Henrique
    4. Zuhlhusni Abdul
    Bâtiment IMAG (206)
    A Flexible Distributed Optimization Scheme with Asynchronous, Scarse, and Sparse Communications, by Franck Iutzeler (LJK)

    Category: Seminars A Flexible Distributed Optimization Scheme with Asynchronous, Scarse, and Sparse Communications, by Franck Iutzeler (LJK)


    June 21, 2018

    We present an asynchronous algorithm for distributed convex optimization when the objective writes a sum of smooth functions, local to each worker, and a non-smooth function. This kind of problem typically appears when learning over distributed data. Unlike many existing methods, our algorithm is adjustable to various levels of communication cost, delays, machines computational power. Moreover, for L1-regularized problems, this algorithm identifies near-optimal sparsity patterns and leverages on it to improve the efficiency of communications.

    Bat. IMAG, 306
  • Pré-soutenances M2R DATAMOVE/POLARIS

    Category: Seminars Pré-soutenances M2R DATAMOVE/POLARIS


    June 22, 2018

    1. Gautheron Flora
    2. Roussillon Benjamin
    3. Dogeas Kostas
    Bâtiment IMAG (206)
    Pré-soutenance de thèse de Benoit Vinot

    Category: Seminars Pré-soutenance de thèse de Benoit Vinot


    June 22, 2018

    Conception d'un système d'information distribué pour la conduite des flexibilités dans un réseau de distribution électrique: modélisation, simulation et implémentation

    Résumé français :

    Le secteur industriel de l'énergie, et les réseaux électriques en particulier, rendent à nos sociétés modernes d'immenses services dont nous ne pouvons plus nous passer. Ils présentent aussi, hélas, un certain nombre de graves inconvénients, notamment en matière d'impact environnemental. Ces inconvénients apparaissent aujourd'hui comme inacceptables; le secteur de l'énergie s'efforce donc actuellement de les amoindrir autant que possible, dans le cadre de ce qu'on appelle la transition énergétique.

    Outre d'indispensables efforts en matière de sobriété et d'efficacité énergétique, deux grands axes d'amélioration se dessinent: d'une part, le remplacement progressif de certains moyens de production
    d'électricité conventionnels par des moyens de production renouvelables; et d'autre part, le transfert de certains usages aujourd'hui non-électriques vers l'électricité --- en particulier en matière de mobilité.

    L'intégration au réseau électrique de ces nouveaux types de dispositifs pose cependant des difficultés techniques considérables, qui motivent depuis le début des années 2000 de nombreux travaux sur le thème de ce que l'on appelle aujourd'hui les "smart grids": des réseaux électriques compatibles avec les exigences de
    la transition énergétique, c'est-à-dire capables d'accueillir massivement les nouveaux types d'usages comme la production photovoltaïque et les bornes de recharge des véhicules électriques, ceci
    notamment grâce à l'utilisation accrue des nouvelles technologies de l'information et de la communication.

    Parmi les difficultés susmentionnées, qui limitent la capacité d'accueil du réseau, figurent les congestions, c'est-à-dire les limites physiques à la puissance que l'on peut faire transiter d'un point à un autre sur une infrastructure donnée. C'est à la gestion des congestions que nos travaux sont consacrés. À ce sujet, la question
    fondamentale est de déterminer par quel enchaînement de mesures, de calcul, de communications et in fine d'actions, on peut passer d'une situation de contrainte sur un réseau de distribution d'électricité, à une situation où cette contrainte a été éliminée par l'action des flexibilités environnantes; autrement dit, en augmentant ou en réduisant judicieusement la production et/ou la consommation locales, et éventuellement en jouant sur d'autres types de leviers.

    L'objet de cette thèse est de participer à l'élaboration des outils conceptuels et informatiques qui nous permettront de répondre à la question fondamentale ci-dessus. Nos travaux portent ainsi sur la question de la modélisation des réseaux de distribution d'électricité "flexibles", et sur l'implémentation concrète des modèles retenus sous forme d'un logiciel de simulation ad hoc, parfaitement adapté à l'étude de ce type de réseaux.

    Abstract :

    The energy sector and the electrical networks in particular, provide great and indispensable services to our modern societies. Unfortunately, they also bring some serious drawbacks, especially with regard to the environment. These drawbacks are becoming more and more unacceptable; that is why the energy sector is
    trying to reduce them as much as possible, in the framework of the so-called energy transition.

    In addition to mandatory efforts in terms of energy efficiency and sobriety, two major directions of improvement have been identified: on the one hand, the progressive replacement of some conventional power plants with renewable production units; and on the other hand, the transfer of several non-electrical usages towards electricity --- in particular in the area of mobility.

    The integration of these new devices into electrical networks raise new technical challenges which, since the early 2000s, have been driving a lot of work about so-called "smart grids": electrical networks compatible with the requirements of the energy transition, ie. able to host new devices like photovoltaic solar panels and charging stations for electric vehicles, notably through the increasing usage of new information and communications technologies.

    Among the difficulties mentioned above, which limit the hosting capacity of the network, there are congestions ie physical constraints limiting the amount of power that may be transmitted through a given infrastructure. Our work is devoted to the management of congestions. The fundamental issue thereon is to define a sequence of decisions, computations, communications and in fine actions that allows to move from a constrained situation on the electrical distribution network, to a situation in which the action of local flexibilities has lifted the constraint; in other words, to a situation where increasing or decreasing local generation and/or consumption, or taking some other control action, relieved the network.

    The aim of this thesis is to contribute to the development of conceptual and computing tools that will allow us to answer the fundamental aforementioned issue. Our work thus deals with the modelling of flexible electrical distribution networks, and with the tangible implementation of selected models in the form of
    ad hoc simulation software, specifically designed for the study of such networks.

    Bâtiment IMAG
    Saint-Martin-d'Hères, 38400
    France
  • Real-Time Scheduling Policy Selection using Machine Learning by Raphael Camargo(Sao Paulo)

    Category: Seminars Real-Time Scheduling Policy Selection using Machine Learning by Raphael Camargo(Sao Paulo)


    June 28, 2018

    We present a real-time scheduling policy selection algorithm, which takes as input the running queue job characteristics and machine states. We evaluated the use of logistic regression and support vector machines to perform the mapping from queue and machine state to selected scheduling policy. The machine learning algorithms are trained and evaluated using simulations configured using HPC platform traces. When selecting among 5 (five) scheduling policies and using SVM with polynomial kernel, we obtained an accuracy above 80%, when compared to the best possible selection. When simulating the online real-time selection of policies for a period of one year, the method had a performance between selecting the best possible selection of policies and random selection, when considering the mean queue waiting time.

    Bâtiment IMAG (206)
  • Minimization of Memory Size in Parallel Computations, by Bruno Gaujal (Polaris).

    Category: Seminars Minimization of Memory Size in Parallel Computations, by Bruno Gaujal (Polaris).


    July 5, 2018

    Given a digital circuit (made of logical gates and registers), or a parallel loop, is it possible to construct a new circuit computing the same function but using less registers?
    We show that the minimal number of registers is the size of a minimal cut in a periodic infinite graph corresponding to an unfolding of the initial circuit.
    We also show when it coincides with the retiming technique introduced by Leiserson and Saxe and when it outperforms optimal retiming.

    Bâtiment IMAG (442)
    Saint-Martin-d'Hères, 38400
    France

Comments are closed.