Decision trees and MPI collective algorithm selection problem Export

@misc{citeulike:3177107, abstract = {Selecting the close-to-optimal collective algorithm based on the parameters of the collective call at run time is an important step in achieving good performance of MPI applications. In this paper, we explore the applicability of C4.5 decision trees to the MPI collective algorithm selection problem. We construct C4.5 decision trees from the measured algorithm performance data and analyze the decision tree properties and expected run time performance penalty. In cases we considered, results show that the C4.5 decision trees can be used to generate a reasonably small and very accurate decision function. For example, the Broadcast decision tree with only 21 leaves was able to achieve a mean performance penalty of 2.08\%. Similarly, combining experimental data for Reduce and Broadcast and generating a decision function from the combined decision trees resulted in less than 2.5 \% relative performance penalty. The results indicate that C4.5 decision trees are applicable to this problem and should be more widely used in this domain. 1}, author = {Fagg, Graham E. and Angskun, Thara and Bosilca, George and Dongarra, Jack J.}, citeulike-article-id = {3177107}, citeulike-linkout-0 = {http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.101.2641}, keywords = {decision-trees}, posted-at = {2008-08-31 21:54:07}, priority = {2}, title = {Decision trees and MPI collective algorithm selection problem}, url = {http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.101.2641}, volume = {2006}, year = {2006} }

TY - ELEC ID - citeulike:3177107 L3 - citeulike-article-id:3177107 N2 - Selecting the close-to-optimal collective algorithm based on the parameters of the collective call at run time is an important step in achieving good performance of MPI applications. In this paper, we explore the applicability of C4.5 decision trees to the MPI collective algorithm selection problem. We construct C4.5 decision trees from the measured algorithm performance data and analyze the decision tree properties and expected run time performance penalty. In cases we considered, results show that the C4.5 decision trees can be used to generate a reasonably small and very accurate decision function. For example, the Broadcast decision tree with only 21 leaves was able to achieve a mean performance penalty of 2.08%. Similarly, combining experimental data for Reduce and Broadcast and generating a decision function from the combined decision trees resulted in less than 2.5 % relative performance penalty. The results indicate that C4.5 decision trees are applicable to this problem and should be more widely used in this domain. 1 TI - Decision trees and MPI collective algorithm selection problem VL - 2006 KW - decision-trees AU - Fagg, Graham AU - Angskun, Thara AU - Bosilca, George AU - Dongarra, Jack PY - 2006/// UR - http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.101.2641 ER -