Performances (informatique)

En informatique, les performances énoncent les indications chiffrées mesurant les possibilités maximales ou optimales d'un matériel, d'un logiciel, d'un système ou d'un procédé technique pour exécuter une tâche donnée. Selon le contexte, les performances incluent les mesures suivantes :

Un faible temps de réponse pour effectuer une tâche donnée
Un débit élevé (vitesse d'exécution d'une tâche)
L'efficience : faible utilisation des ressources informatiques : processeur, mémoire, stockage, réseau, consommation électrique, etc.
Une haute disponibilité du système ou de l'application
Une bande passante élevée ou une faible latence
Le passage à l'échelle (scalabilité)
La capacité d'un canal de communication
La performance du stockage informatique

Les performances sont établies par la mesure lors d'une activité appelée benchmarking (en français, étalonnage), grâce à des tests de performance. Le test de performance consiste à définir le processus/système/matériel que l'on veut

Decentralized in-order execution of a sequential task-based code for shared-memory architectures

Charly Nicolas Lucien Castes

The hardware complexity of modern machines makes the design of adequate programming models crucial for jointly ensuring performance, portability, and productivity in high-performance computing (HPC). Sequential task-based programming models paired with advanced runtime systems allow the programmer to write a sequential algorithm independently of the hardware architecture in a productive and portable manner, and let a third party software layer -the runtime system- deal with the burden of scheduling a correct, parallel execution of that algorithm to ensure performance. Many HPC algorithms have successfully been implemented following this paradigm, as a testimony of its effectiveness. Developing algorithms that specifically require fine-grained tasks along this model is still considered prohibitive, however, due to per-task management overhead [1], forcing the programmer to resort to a less abstract, and hence more complex "task+X" model. We thus investigate the possibility to offer a tailored execution model, trading dynamic mapping for efficiency by using a decentralized, conservative in-order execution of the task flow, while preserving the benefits of relying on the sequential taskbased programming model. We propose a formal specification of the execution model as well as a prototype implementation, which we assess on a shared-memory multicore architecture with several synthetic workloads. The results show that under the condition of a proper task mapping supplied by the programmer, the pressure on the runtime system is significantly reduced and the execution of fine-grained task flows is much more efficient.

IEEE COMPUTER SOC2022

Étude du tomographe de haute résolution pour petits animaux ClearPET par la méthode de Monte Carlo

Martin Rey

The Lausanne ClearPET demonstrator is one of the new generation of high resolution small animal PET scanners. A high resolution PET scanner aims to maximize the signal-to-noise ratio measured in pixels for a given time without compromising spatial resolution. In order to achieve it, ClearPET scanners are based on phoswich technology : two different crystals (LSO and LuYAP) are aligned one behind the other and coupled to the same channel of a multichannel photo-detector. Depth-of-interaction is determined by a pulse shape analysis. To improve the prototype design, a Monte Carlo simulation toolkit dedicated to emission tomography – GATE – was created. The accurate description of time-dependent phenomena such as source or detector movement and source decay kinetics represents the most important feature of this software. The first part of this work presents the demonstrator built in Lausanne, mainly the DAQ process and the libraries for the data treatment, and the GATE functionality. In the second part, the measurements obtained with the ClearPET demonstrator combined with simulations are presented. The simulations allow estimation of the performance of a final scanner and refinement of the detector head design. Measurements as well as simulations give a spatial resolution of 1.3 mm on the scanner axis and 2.5 mm at 4 cm from the axis. Temporal resolution for two modules with the same sampling phase is about 4.3 ns for LSO and 4.9 ns for LuYAP. For a standard acquisition, the energy resolution at 511 keV is about 31 ± 4 % for LSO and 33 ± 8 % for LuYAP. The peaks at full energy before calibration are about 480 ± 50 keV1 for LSO and 470 ± 40 keV1 for LuYAP. These variations, coupled with hardware threshold, are one of the main reasons for the sensitivity and count rate performance limitations. A sensitivity of 4.37 ± 0.05 is estimated for a full ring design with four rings of detector modules. The large systematic errors are induced by the variability previously mentioned. ------------------------------ 1 As spectra are not normalized, keV unities are inappropriate. Channel numbers are more accurate.

EPFL2007

Étude du tomographe de haute résolution pour petits animaux ClearPET par la méthode de Monte Carlo

Martin Rey

EPFL2007

Prédiction de performance et optimisation d'un plan de voilure

Decentralized in-order execution of a sequential task-based code for shared-memory architectures

Étude du tomographe de haute résolution pour petits animaux ClearPET par la méthode de Monte Carlo

Prédiction de performance et optimisation d'un plan de voilure

Étude du tomographe de haute résolution pour petits animaux ClearPET par la méthode de Monte Carlo

Decentralized in-order execution of a sequential task-based code for shared-memory architectures