Minas: Memory Affinity Management Framework

Christiane Pousa Ribeiro 1 Jean-François Méhaut 2
2 MESCAL - Middleware efficiently scalable
Inria Grenoble - Rhône-Alpes, LIG - Laboratoire d'Informatique de Grenoble
Abstract : In this document, we introduce Minas, a memory affinity management framework for cache-coherent NUMA Non-Uniform Memory Access) platforms, which provides either explicit memory affinity management or automatic one with efficiency and architecture abstraction for numerical scientic applications. The explicit tuning is based on an API called MAi (Memory Affinity interface) which provides simple functions to manage allocation and data placement using an extensive set of memory policies. An automatic tuning mechanism is provided by the preprocessor named MApp (Memory Anity preprocessor). MApp analyses both the application source code and the target cache-coherent NUMA platform characteristics in order to automatically apply MAi functions at compile time. Minas efficiency and architecture abstraction have been evaluated on two cache-coherent NUMA platforms using three numerical scientic HPC applications. The results have shown signicant gains when compared to other solutions available on Linux (First-touch, libnuma and numactl).
Liste complète des métadonnées

Cited literature [15 references]  Display  Hide  Download

https://hal.inria.fr/inria-00421546
Contributor : Christiane Pousa Ribeiro <>
Submitted on : Monday, November 2, 2009 - 7:00:03 AM
Last modification on : Thursday, October 11, 2018 - 8:48:02 AM
Document(s) archivé(s) le : Tuesday, June 15, 2010 - 10:24:28 PM

File

RT-7051.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00421546, version 1

Collections

Citation

Christiane Pousa Ribeiro, Jean-François Méhaut. Minas: Memory Affinity Management Framework. [Research Report] RR-7051, INRIA. 2009. ⟨inria-00421546⟩

Share

Metrics

Record views

301

Files downloads

203