Latest news:
Sept 10, 2009:
version 0.1 released!
ACBEA started as a master thesis proposed by Sebastien Varrette
that Dominic Dunlop accepted to take care (also under the supervision
of Pascal Bouvry).
The initial approach was directly inspired by Acovea (Analysis of Compiler Options via Evolutionary Algorithm), an open-source framework that investigate the optimum combination of command-line flags for a compiler (GCC typically) given a given goal (executable size, speed etc.). The exploration of the possible flag combination is guided by an evolutionary algorithm.
In a first work, a wrapper to Acovea has been designed to investigate the optimal parameters to be given to HPL in order to maximize the benchmark result.
This first study has been validated over the clusters present at the University of Luxembourg and is reported in [DVB_SPECTS08].
Later on, it was proposed to analyse the impact of the underlying evolutionary algorithm and to design an integrated tool that manage in a nearly automatic way the process of cluster benchmarking. The analysis and the resulting framework, called ACBEA, together with the validation over various clusters in Grid'5000, is presented in [DVB_PPAM09].
The initial approach was directly inspired by Acovea (Analysis of Compiler Options via Evolutionary Algorithm), an open-source framework that investigate the optimum combination of command-line flags for a compiler (GCC typically) given a given goal (executable size, speed etc.). The exploration of the possible flag combination is guided by an evolutionary algorithm.
In a first work, a wrapper to Acovea has been designed to investigate the optimal parameters to be given to HPL in order to maximize the benchmark result.
This first study has been validated over the clusters present at the University of Luxembourg and is reported in [DVB_SPECTS08].
Later on, it was proposed to analyse the impact of the underlying evolutionary algorithm and to design an integrated tool that manage in a nearly automatic way the process of cluster benchmarking. The analysis and the resulting framework, called ACBEA, together with the validation over various clusters in Grid'5000, is presented in [DVB_PPAM09].
| [DVB_PPAM09] | D. Dunlop, S. Varrette and P. Bouvry. Deskilling HPL - Using an Evolutionary Algorithm to Automate Cluster Benchmarking. LNCS, In: Proc. of 8th Intl. conf. on Parallel Processing and Applied Mathematics (PPAM 2009) , Springer Verlag publishing, Wroclaw, Polan, Sept 2009. |
| [DVB_SPECTS08] | D. Dunlop, S. Varrette and P. Bouvry.. On the Use of a Genetic Algorithm in High Performance Computer Benchmark Tuning. In: Proc. of the IEEE Int. Symposium on Performance Evaluation of Computer and Telecommunication Systems (SPECTS'08), pages 105--113, IEEE Xplore publishing, Edinburgh, UK, June 2008. |
| Deskilling HPL - Using an Evolutionary Algorithm to Automate Cluster Benchmarking. by D. Dunlop, S. Varrette and P. Bouvry (in 2009). | ![[urlpdf]](images/urlpdf.gif){kind=small}
| ||||||||||||||||||||||||||||||
| (top) @InProceedings{DVB_PPAM09,
| |||||||||||||||||||||||||||||||
| On the Use of a Genetic Algorithm in High Performance Computer Benchmark Tuning. by D. Dunlop, S. Varrette and P. Bouvry. (in 2008). |
| ||||||||||||||||||||||||||||||
| (top) @InProceedings{DVB_SPECTS08,
| |||||||||||||||||||||||||||||||
