The main result is a new original algorithm: CONDOR ("COnstrained, Non-linear, Direct, parallel Optimization using trust Region method for high-computing load, noisy functions"). The aim of this algorithm is to find the minimum x* of an objective function F(x) (x is a vector whose dimension is between 1 and 150) using the least number of function evaluations of F(x). It is assumed that the dominant computing cost of the optimization process is the time needed to evaluate the objective function F(x) (One evaluation can range from 2 minutes to 2 days). The algorithm will try to minimize the number of evaluations of F(x), at the cost of a huge amount of routine work. CONDOR is a derivate-free optimization tool (i.e., the derivatives of F(x) are not required. The only information needed about the objective function is a simple method (written in Fortran, C++,...) or a program (a Unix, Windows, Solaris,... executable) which can evaluate the objective function F(x) at a given point x. The algorithm has been specially developed to be very robust against noise inside the evaluation of the objective function F(x). This hypotheses are very general, the algorithm can thus be applied on a vast number of situations. CONDOR is able to use several CPU's in a cluster of computers. Different computer architectures can be mixed together and used simultaneously to deliver a huge computing power. The optimizer will make simultaneous evaluations of the objective function F(x) on the available CPU's to speed up the optimization process. The experimental results are very encouraging and validate the quality of the approach: CONDOR outperforms many commercial, high-end optimizer and it might be the fastest optimizer in its category (fastest in terms of number of function evaluations). When several CPU's are used, the performances of CONDOR are currently unmatched (may 2004). CONDOR has been used during the METHOD project to optimize the shape of the blades inside a Centrifugal Compressor (METHOD stands for Achievement Of Maximum Efficiency For Process Centrifugal Compressors THrough New Techniques Of Design). In this project, the objective function is based on a 3D-CFD (computation fluid dynamic) code which simulates the flow of the gas inside the compressor.

L’évolution des supercalculateurs est à la source de défis logiciels et architecturaux. Dans la quête de puissance de calcul, l’interdépendance des éléments du processus de simulation devient de plus en plus impactante et requiert de nouvelles approches. Cette thèse se concentre sur le développement logiciel et particulièrement sur l’observation des programmes parallèles s’exécutant sur des milliers de cœurs. Dans ce but, nous décrivons d’abord le processus de développement de manière globale avant de présenter les outils existants et les travaux associés. Dans un second temps, nous détaillons notre contribution qui consiste d’une part en des outils de débogage et profilage par prise de traces, et d’autre part en leur évolution vers un couplage en ligne qui palie les limitations d’entrées–sorties. Notre contribution couvre également la synchronisation des horloges pour la prise de traces avec la présentation d’un algorithme de synchronisation probabiliste dont nous avons quantifié l’erreur. En outre, nous décrivons un outil de caractérisation machine qui couvre l’aspect MPI. Un tel outil met en évidence la présence de bruit aussi bien sur les communications de type point-à-point que de type collective. Enfin, nous proposons et motivons une alternative à la collecte d’événements par prise de traces tout en préservant la granularité des événements et un impact réduit sur les performances, tant sur le volet utilisation CPU que sur les entrées–sorties / Supercomputers’ evolution is at the source of both hardware and software challenges. In the quest for the highest computing power, the interdependence in-between simulation components is becoming more and more impacting, requiring new approaches. This thesis is focused on the software development aspect and particularly on the observation of parallel software when being run on several thousand cores. This observation aims at providing developers with the necessary feedback when running a program on an execution substrate which has not been modeled yet because of its complexity. In this purpose, we firstly introduce the development process from a global point of view, before describing developer tools and related work. In a second time, we present our contribution which consists in a trace based profiling and debugging tool and its evolution towards an on-line coupling method which as we will show is more scalable as it overcomes IOs limitations. Our contribution also covers our time-stamp synchronisation algorithm for tracing purposes which relies on a probabilistic approach with quantified error. We also present a tool allowing machine characterisation from the MPI aspect and demonstrate the presence of machine noise for both point to point and collectives, justifying the use of an empirical approach. In summary, this work proposes and motivates an alternative approach to trace based event collection while preserving event granularity and a reduced overhead

The main result is a new original algorithm: CONDOR ("COnstrained, Non-linear, Direct, parallel Optimization using trust Region method for high-computing load, noisy functions"). The aim of this algorithm is to find the minimum x* of an objective function F(x) (x is a vector whose dimension is between 1 and 150) using the least number of function evaluations of F(x). It is assumed that the dominant computing cost of the optimization process is the time needed to evaluate the objective function F(x) (One evaluation can range from 2 minutes to 2 days). The algorithm will try to minimize the number of evaluations of F(x), at the cost of a huge amount of routine work. CONDOR is a derivate-free optimization tool (i.e. the derivatives of F(x) are not required. The only information needed about the objective function is a simple method (written in Fortran, C++,) or a program (a Unix, Windows, Solaris, executable) which can evaluate the objective function F(x) at a given point x. The algorithm has been specially developed to be very robust against noise inside the evaluation of the objective function F(x). This hypotheses are very general, the algorithm can thus be applied on a vast number of situations. CONDOR is able to use several CPU's in a cluster of computers. Different computer architectures can be mixed together and used simultaneously to deliver a huge computing power. The optimizer will make simultaneous evaluations of the objective function F(x) on the available CPU's to speed up the optimization process. The experimental results are very encouraging and validate the quality of the approach: CONDOR outperforms many commercial, high-end optimizer and it might be the fastest optimizer in its category (fastest in terms of number of function evaluations). When several CPU's are used, the performances of CONDOR are currently unmatched (may 2004). CONDOR has been used during the METHOD project to optimize the shape of the blades inside a Centrifugal Compressor (METHOD stands for Achievement Of Maximum Efficiency For Process Centrifugal Compressors THrough New Techniques Of Design). In this project, the objective function is based on a 3D-CFD (computation fluid dynamic) code which simulates the flow of the gas inside the compressor. / Doctorat en sciences appliquées / info:eu-repo/semantics/nonPublished