Using Automated Performance Modeling to Find Scalability Bugs in Complex Codes
By A. Calotoiu, T. Hoefler, M. Poke, and F. Wolf
Published in Proceedings of the ACM/IEEE Conference on Supercomputing (SC13), Denver, CO, USA
Extra-P is an automatic performance-modeling tool that supports the user in the identification of scalability bugs. A scalability bug is a part of the program whose scaling behavior is unintentionally poor, that is, much worse than expected.
Extra-P uses measurements of various performance metrics at different processor configurations as input to represent the performance of code regions (including their calling context) as a function of the number of processes. All it takes to search for scalability issues even in full-blown codes is to run a manageable number of small-scale performance experiments, launch Extra-P, and compare the asymptotic or extrapolated performance of the worst instances to the expectations. Besides the number of processes, it is also possible to consider other parameters such as the input problem size.
Extra-P generates not only a list of potential scalability bugs but also human-readable models for all performance metrics available such as floating-point operations or bytes sent by MPI calls that can be further analyzed and compared to identify the root causes of scalability issues.This software is free but copyrighted
|© 1998-2015||Forschungszentrum Jülich GmbH, Germany|
|© 2009-2015||German Research School for Simulation Sciences GmbH, Jülich/Aachen, Germany|
|© 2015-2016||Technische Universität Darmstadt, Darmstadt, Germany|
|© 2014-2015||Lawrence Livermore National Laboratory, Livermore, California, USA|
|Extra-P 2.0b package for download|
MD5sum : 96a830fa79b7486fce9b7065adc7dd7c
Extra-P has been successfully tested on the following platforms:
- Linux (x86_64)
- Mac OS X (x86_64) (10.8 and younger)
Please report success/failure on other platforms to the Extra-P development team.