Logo des Repositoriums

Auflistung nach Autor:in "Heuveline, Vincent"

1 - 2 von 2
  • Zeitschriftenartikel
    Development and implementation of a temperature monitoring system for HPC systems
    (PARS-Mitteilungen: Vol. 34, Nr. 1, 2017) Baumann, Martin; Gebhart, Fabian; Mattes, Oliver; Nikas, Sotirios; Heuveline, Vincent
    In the context of high-performance computing (HPC), the removal of released heat is one challenging topic due to the continuously increasing density of computing power. A temperature monitoring system provides insight into the heat development of an HPC cluster. The effectiveness of this is directly related to the number of sensors, their placing and the accuracy of the temperature measurements. Monitoring is important not only to investigate the efficiency of the cooling system for purposes of detecting defective operation of the HPC system, but also to improve the cooling of the servers and by this the achievable performance. The main purpose of a fine-grained and unified temperature monitoring is the possibility to optimize the applications and their execution regarding the temperature spreading on HPC systems. Based on this, we present a highly flexible and scalable – in terms of cable length and number of sensors – and at the same time budget-friendly monitoring infrastructure. It is based on low-cost components such as Raspberry Pi as monitoring client and a setup using the DS18B20 digital thermometer as temperature sensor. Focus is given on the selection of adequate temperature sensors and we explain in detail how the sensors are assembled and the quality assurance is done before these are used in the monitoring setup.
  • Zeitschriftenartikel
    Energy-aware mixed precision iterative refinement for linear systems on GPU-accelerated multi-node HPC clusters
    (PARS-Mitteilungen: Vol. 32, Nr. 1, 2015) Wlotzka, Martin; Heuveline, Vincent
    Modern high-performance computing systems are often built as a cluster of interconnected compute nodes, where each node is built upon a hybrid hardware stack of multi-core processors and many-core accelerators. To efficiently use such systems, numerical methods must embrace the different levels of parallelism from the coarse-grained distributed memory cluster level to the fine-grained shared memory node level parallelism. Synchronization requirements of numerical methods may diminish parallel performance and result in increased energy consumption. We investigate block-asynchronous iteration methods in combination with mixed precision iterative refinement to address this issue. We depict our implementation for multi-node distributed systems using MPI with a hybrid node level parallelization for multi-core CPUs using OpenMP and multiple CUDAcapable accelerators. Our numerical experiments are based on a linear system arising from the finite element discretization of the Poisson equation. We present energy and runtime measurements for a quad-CPU and dual-GPU test system. We achieve runtime and energy savings of up to 70% for block-asynchronous GPU-accelerated iteration using mixed precision compared to CPU-only computation. We also encounter configurations where the CPU-only computation is advantageous over the GPU-accelerated method.