Fast and Faithful Performance Prediction of MPI Applications: the HPL Case Study
作者： Tom Cornebize , Arnaud Legrand , Franz C. Heinrich
DOI: 10.1109/CLUSTER.2019.8891011
关键词:
摘要: Finely tuning MPI applications (number of processes, granularity, collective operation algorithms, topology and process placement) is critical to obtain good performance on supercomputers. With a rising cost modern supercomputers, running parallel at scale solely optimize their extremely expensive. Having inexpensive but faithful predictions expected could be great help for researchers system administrators. The methodology we propose captures the complexity adaptive by emulating code while skipping insignificant parts. We demonstrate its capability with High Performance Linpack (HPL), benchmark used rank supercomputers in TOP500 which requires careful tuning. explain (1) how both extended SimGrid’s SMPI simulator slightly modified open-source version HPL allow fast emulation single commodity server supercomputer (2) model different components (network, BLAS, …) system. show that modeling spatial temporal node variability allows us within few percents real experiments (see Figure 1).
sci-hub.se 参考文章(23)
Piotr Luszczek, Jack Dongarra, Reducing the time to tune parallel dense linear algebra routines with partial execution and performance modeling parallel processing and applied mathematics. pp. 730- 739 ,(2011) , 10.1007/978-3-642-31464-3_74
Daniel Balouek, Alexandra Carpen Amarie, Ghislain Charrier, Frédéric Desprez, Emmanuel Jeannot, Emmanuel Jeanvoine, Adrien Lèbre, David Margery, Nicolas Niclausse, Lucas Nussbaum, Olivier Richard, Christian Perez, Flavien Quesnel, Cyril Rohr, Luc Sarzyniec, Adding Virtualization Capabilities to the Grid’5000 Testbed international conference on cloud computing and services science. ,vol. 367, pp. 3- 20 ,(2012) , 10.1007/978-3-319-04519-1_1
James Franklin, The elements of statistical learning : data mining, inference,and prediction The Mathematical Intelligencer. ,vol. 27, pp. 83- 85 ,(2005) , 10.1007/BF02985802
R. F. Bird, S. A. Wright, D. A. Beckingsale, S. A. Jarvis, Performance modelling of magnetohydrodynamics codes EPEW'12 Proceedings of the 9th European conference on Computer Performance Engineering. pp. 197- 209 ,(2012) , 10.1007/978-3-642-36781-6_14
Bettina Grün, Friedrich Leisch, FlexMix Version 2: Finite Mixtures with Concomitant Variables and Varying and Constant Parameters Journal of Statistical Software. ,vol. 28, pp. 1- 35 ,(2008) , 10.18637/JSS.V028.I04
Henri Casanova, Frédéric Desprez, George S. Markomanolis, Frédéric Suter, Simulation of MPI applications with time-independent traces Concurrency and Computation: Practice and Experience. ,vol. 27, pp. 1145- 1168 ,(2015) , 10.1002/CPE.3278
Laura Carrington, Michael A. Laurenzano, Ananta Tiwari, Inferring Large-Scale Computation Behavior via Trace Extrapolation 2013 IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum. pp. 1667- 1674 ,(2013) , 10.1109/IPDPSW.2013.137
Pedro Velho, Lucas Mello Schnorr, Henri Casanova, Arnaud Legrand, On the validity of flow-level tcp network models for grid and cloud simulations ACM Transactions on Modeling and Computer Simulation. ,vol. 23, pp. 23- ,(2013) , 10.1145/2517448
Christian Engelmann, Scaling to a million cores and beyond: Using light-weight simulation to understand the challenges ahead on the road to exascale Future Generation Computer Systems. ,vol. 30, pp. 59- 65 ,(2014) , 10.1016/J.FUTURE.2013.04.014
Luigi Genovese, Alexey Neelov, Stefan Goedecker, Thierry Deutsch, Seyed Alireza Ghasemi, Alexander Willand, Damien Caliste, Oded Zilberberg, Mark Rayson, Anders Bergman, Reinhold Schneider, Daubechies wavelets as a basis set for density functional pseudopotential calculations Journal of Chemical Physics. ,vol. 129, pp. 014109- 014109 ,(2008) , 10.1063/1.2949547