F. Cappello and O. Richard. Performance characteristics of a network of commodity multiprocessors for the nas benchmarks using a hybrid memory model. In Proceedings of PACT'99. Also available at: http://www.lri.fr/ fci/goinfreWWW/PACT99.ps, 1999. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Performance Evaluation of the Omni OpenMP Compiler - Kusano, Satoh, Sato (2000) (4 citations) (Correct)
.... related to OpenMP, for example, research to execute an OpenMP program on top of the Distributed Shared Memory(DSM) environment on a network of workstations[7] and the investigation of a parallel programming model based on the MPI and the OpenMP to utilize the memory hierarchy of an SMP cluster[9]. Several projects, including OpenMP ARB, have stated the intention to develop an OpenMP benchmark program, though Microbenchmarks[6] is the only one available now. 7 Conclusions This paper presented an overview of the Omni OpenMP compiler and an evaluation of its performance. The Omni consists ....
F. Cappello and O. Richard, "Performance characteristics of a network of commodity multiprocessors for the NAS benchmarks using a hybrid memory model", PACT '99, pp.108-116, Oct., 1999.
Understanding performance of SMP clusters running MPI.. - Cappello, Richard, Etiemble Self-citation (Cappello) (Correct)
....the two approaches can be directly used. In [1] we have presented some results on a 8 node clusters for a HMM version which used MPI for communication between nodes and OpenMP for multithreading inside each node. Performance scaling with successive generation of IA32 processors was considered. In [2], results have been presented for a Myrinet cluster of 36 2 way PCs. HMM is supposed to deliver better performance as it matches the hardware features. However, the programming model is more complicated and MPI programs cannot be directly Preprint submitted to Elsevier Preprint 25 February 2000 ....
....Comm) is a shortcut for Computation (resp. Communication) 1 and 2 refers to 1 way and 2 way nodes. 2 way nodes is greater than for 1 way nodes, which means that the speed up is less than 2. Computational speedup also degrades when switching from 2way to 4 way nodes. We have demonstrated in [2] that the performance of the Pentium II system bus is the bottleneck. 4.2 Communication times Comparing left scale and right scale for each benchmark indicates the relative impact of communication times on the overall execution time. CG and FT, for which parallel eciency is lower, both have ....
F. Cappello and O. Richard. Performance characteristics of a network of commodity multiprocessors for the nas benchmarks using a hybrid memory model. In Proceedings of PACT'99. Also available at: http://www.lri.fr/ fci/goinfreWWW/PACT99.ps, 1999.
Investigating the performance of two programming models .. - Cappello, Richard.. (2000) (2 citations) Self-citation (Cappello) (Correct)
....between MPI processes and intra node parallelism within each MPI process. The complete intra node parallelization process for the NAS parallel benchmark has been previously described [12] A detailed analysis of the performance of this approach for the NAS parallel benchmark is presented in [13]. OpenMP parallelization cannot be applied to the whole MPI code including the communication section. Reasons are two folds. First OpenMP environments running at system level map OpenMP threads on top of system threads. We don t know a thread aware MPI environment. Second, parallelizing the ....
F. Cappello and O. Richard. Performance characteristics of a network of commodity multiprocessors for the nas benchmarks using a hybrid memory model. In Proceedings of PACT'99. also available at: http://www.lri.fr/ fci/goinfreWWW/PACT99.ps, 1999.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC