A complete and efficient CUDA-sharing solution for HPC clusters

Abstract

Registro de acceso restringido Este recurso no está disponible en acceso abierto por política de la editorial. No obstante, se puede acceder al texto completo desde la Universitat Jaume I o si el usuario cuenta con suscripción. Registre d'accés restringit Aquest recurs no està disponible en accés obert per política de l'editorial. No obstant això, es pot accedir al text complet des de la Universitat Jaume I o si l'usuari compta amb subscripció. Restricted access item This item isn't open access because of publisher's policy. The full--text version is only available from Jaume I University or if the user has a running suscription to the publisher's contents.

Multi-tenant virtual GPUs for optimising performance of a financial risk application

“…For copies using pageable memory rCUDA even performs better than CUDA; this has been previously reported, which is due to the use of an efficient pipelined communication between rCUDA clients and servers based on the use of internal and pre-allocated pinned memory buffers [8]. rCUDA and DS-CUDA support InfiniBand Verbs API and therefore have access to large bandwidths which are available on this interconnect.…”

“…The rCUDA framework [8] is binary compatible with CUDA 6.5 and implements the entire CUDA Runtime and Driver APIs (with the exception of graphics functions). It provides support for the libraries included within CUDA, such as cuBLAS or cuFFT.…”

“…Numerous challenges arise when multiple GPUs are shared across a cluster for an application, of which three are considered in this paper. The challenges are addressed in this paper by exploring remote CUDA (rCUDA) [8], a GPU virtualisation framework, for improving the performance of a real-world case study employed in the financial industry. The application typically runs in a cluster environment, but can hugely benefit from GPU acceleration for deriving important risk metrics in real-time.…”

A Performance Comparison of CUDA Remote GPU Virtualization Frameworks

“…Actually, for copies using pageable memory rCUDA also performs better than CUDA. This is a well-known effect thoroughly described in previous works on rCUDA [6] and is due to the use of an efficient pipelined communication based on the use of internal preallocated pinned memory buffers. Nevetherless, the rCUDA performance shown in the figure is a noticeable improvement over other previously published results.…”

“…In addition, GPU utilization is usually low. In this context, remote GPU virtualization frameworks have been recently created to overcome these drawbacks, such as VGPU [9], GridCuda [4], DS-CUDA [5], GVirtuS [2], vCUDA [8], GViM [3], rCUDA [6], and Shadowfax II [7].…”