Supercomputing Laboratory

The Abstract Data and Communication Library (ADCL) is an adaptive communication library optimizing application level collective communication operations at runtime. The library provides for a given communication pattern a large number of implementations and incorporates a runtime selection logic in order to choose the implementation leading to the highest performance. In this paper, we demonstrate for the first time, how an application utilizing ADCL is deployed on a wide range of HPC architectures, including an IBM Blue Gene, an NEC SX8, an IBM Power PC cluster using an IBM Federation Switch, an AMD Opteron cluster utilizing an 4xInfiniBand and a Gigabit Ethernet network, and an Intel EM64T cluster using a hierarchical Gigabit Ethernet network with reduced uplink bandwidth. We demonstrate, how different implementations for the three dimensional neighborhood communication lead to the minimal execution time of the application on different architectures. ADCL gives the user the advantage of having to maintain only a single version of the source code and still have the ability to achieve close to optimal performance for the application on all architectures.

Le volume élevé des Prêts Non Performants (PNP) reste toujours le problème le plus important dans le secteur bancaire tunisien. La classification des banques selon leur niveau des PNP semble alors cruciale pour les autorités de supervision locales afin de prendre les actions correctives adéquates. Dans cette étude, nous comparons trois méthodes de classification différentes, à savoir: les réseaux de neurones, les machines à vecteurs supports multiclasses et le classifieur bayésien sous l'hypothèse de normalité. Nous utilisons des données spécifiques aux dix banques commerciales tunisiennes collectées sur une période de sept ans, de 2000 à 2006. Il s'agit de vingt huit facteurs de risque appartenant au rating CAMELS. Ces facteurs sont ordonnés selon leur importance et sélectionnés en utilisant des algorithmes de sélection de variables.

This paper presents methods of banks discrimination according to the rate of NonPerforming Loans (NPLs), using Gaussian Bayes models and different approaches of multiclass Support Vector Machines (SVM). This classification problem involves many irrelevant variables and comparatively few training instances. New variable selection strategies are proposed. They are based on Gaussian marginal densities for Bayesian models and ranking scores derived from multiclass SVM. The results on both toy data and real-life problem of banks classification demonstrate a significant improvement of prediction performance using only a few variables. Moreover, Support Vector Machines approaches are shown to be superior to Gaussian Bayes models.

Emerging computing systems have a wide variety of hardware and software components influencing the performance of parallel applications, presenting end-users with a (nearly) unique execution environment on each parallel machine. One of the big challenges of High Performance Computing is therefore to develop portable and efficient codes for any execution environment. The Abstract Data and Communication Library (ADCL) is a self-optimizing runtime communication library aiming at providing the highest possible performance for application level communication operations. The library provides for a given communication pattern a large number of implementations and incorporates a runtime selection logic. This selection aims at adaptively choosing the best performing implementation on the current platform and for the given problem. In this paper, we present a recent enhancement to the library which introduces the capability of utilizing information from previous executions in order to minimize the overhead of the runtime selection logic which mainly stems from testing underperforming implementations. We introduce the notion of similar problems by using a proximity measure for a given operation. The approach is evaluated for the n-dimensional neighborhood communication on different platforms and for a large range of different problems.

Clustered computing environments, although becoming the predominant high-performance computing platform of choice, continue to grow in complexity. It is relatively easy to achieve good performance with real-world MPI applications on such platforms, but obtaining the best possible MPI performance is still an extremely difficult task, requiring painstaking tuning of all levels of the hardware and software in the system. The Open Tool for Parameter Optimization (OTPO) is a new framework designed to aid in the optimization of one of the key software layers in high performance computing: Open MPI. OTPO systematically tests large numbers of combinations of Open MPI's run-time tunable parameters for common communication patterns and performance metrics to determine the "best" set for a given platform. This paper presents the concept, some implementation details and the current status of the tool, as well as an example optimizing InfiniBand message passing latency by Open MPI.