Hands-on Session: Performance Analysis and Optimization

Instructor: Phil Blood
Pittsburgh Supercomputing Center
blood@psc.edu

Instructor: Christian Feld
Jülich Supercomputing Centre
c.feld@fz-juelich.de

The session gives an introduction into performance engineering of serial and parallel applications. It will cover open-source tools that provide a practical basis for portable performance analysis of application execution, covering both profiling and tracing. The session will be delivered as a series of presentations with associated hands-on practical exercises using the XSEDE's Bridges supercomputer. The analysis of provided example codes will be used to guide the class through the relevant steps of a measurement and analysis workflow cycle.

Preparations

Please install the required software (ParaProf and Cube) as describe here: Install required Software and Data.

Slides

1. Blood_Performance_Engineering_IntlSchool2017.pdf
   
2. IHPCSS_01_HandsOn_BTMZ_Reference.pdf
3. IHPCSS_02_Score-P_Intro_HandsOn.pdf
4. Blood_Paraprof_IntlSchool2017.pdf
   
5. IHPCSS_03_Tracing_Intro.pdf, from slide 20 on 
   
6. IHPCSS_04_HandsOn_Scoring_and_Trace_Collection.pdf
7. IHPCSS_05_Cube.pdf
   see also Performance Properties
   
8. IHPCSS_06_VI-HPS_Tools.pdf
   

Measurements

Download this PAPI experiment for the Paraprof hands-on session:

• PAPI-experiments (tar.gz, zip)

In case, for whatever reason, we cannot collect measurements on Bridges or Comet, please use these prepared measurements:

• bridges-experiments (tar.gz, zip)
  
• comet-experiments (tar.gz)

Programming challenge Score-P measurement results as showed during the session, just for MPI:

• challenge-scorep-MPI.tar.gz
  

Example code

The example code is available on Bridges and Comet, this version is just for reference:

• NPB3.3-MZ-MPI.tar.gz
  

Philip Blood received his Ph.D. in Bioengineering from the University of Utah where he utilized massively parallel molecular dynamics simulations to study how proteins remodel cellular membranes. In 2007, after extensively employing NSF supercomputers to conduct his research, Philip joined the Pittsburgh Supercomputing Center where he works as a Senior Computational Scientist. He is actively engaged in the NSF XSEDE project, working with scientists in the fields of computational chemistry, biophysics, bioinformatics, and various other disciplines to advance science through supercomputing. Philip runs the Anton project at PSC providing researchers access to a special-purpose supercomputer for biomolecular simulation. As a PI of the National Center for Genome Analysis Support, he is also focused on enabling large-scale, accessible genomics analysis.

Christian Feld has been active in the field of computer simulation and performance analysis since his time as a student at Cologne University from where he received his Diploma in Physics in 2004. After a year working as a consultant and software engineer for the Münster University of Applied Sciences, he moved to PTV AG in Karlsruhe where he joined the traffic simulation development team. From 2009 on he works as a scientific staff member at the Jülich Supercomputing Centre (JSC), reinforcing JSC’s cross-sectional team “Performance Analysis”. This gave him the opportunity to visit University of Oregon as courtesy research assistant and becoming visiting scientist at RIKEN’s Advanced Institute for Computational Science, Programming Environment Research Team. His work is primarily focused within projects developing Score-P, the next generation measurement infrastructure for the performance analysis tools Scalasca, Vampir, TAU and Periscope.