LogP: towards a realistic model of parallel computation
PPOPP '93 Proceedings of the fourth ACM SIGPLAN symposium on Principles and practice of parallel programming
Analyzing scheduling policies using Dimemas
Parallel Computing - Special double issue on environment and tools for parallel scientific computing
Performance Profiling and Analysis of DoD Applications Using PAPI and TAU
DOD_UGC '05 Proceedings of the 2005 Users Group Conference on 2005 Users Group Conference
Scientific Programming - High Performance Computing with the Cell Broadband Engine
Fast pattern-specific routing for fat tree networks
ACM Transactions on Architecture and Code Optimization (TACO)
Hi-index | 0.00 |
The MareNostrum is a large PowerPC cluster installed at the Barcelona Supercomputing Center and was ranked fourth in the November 2005 Top500 list. This article presents an evaluation of the MareNostrum from the individual components to the final performance delivered for real applications. Beyond just reporting the achieved performance metrics, the paper aims to develop models of the machine and applications. The results of microbenchmarks and parameter fits based on simulations with Dimemas are used to generate the models. Such characterizations prove very useful to properly predict and understand the observed behavior and extrapolate expectations.