Event Detail

Registration

Sorry, the registration period for this event is over.

2011 Course/Workshop: Scalable Performance Analysis Tools for HPC Applications

 

The HP2C platform announces a 2-day intensive course focused on the use of performance analysis tools to analyze application executions on massively parallel processing (MPP) systems. One day is reserved for the Scalasca toolset (www.scalasca.org), presented by Dr. Brian Wylie of Jülich Supercomputing Center, the other is reserved for TAU performance system (http://tau.uoregon.edu), presented by Dr. Sameer Shende from ParaTools. A course syllabus will be posted subsequently.

Registration deadline: January 10, 2011

Please contact wsawyer(at)cscs.ch for further technical information and apinna(at)cscs.ch for logistical information.

 

Instructors

 

Dates

Dr. Brian Wylie from Jülich Supercomputing Center
Dr. Sameer Shende from ParaTools.


17.-18.01.2011

Venue

ETH Zurich, Room HG E 23, Hauptgebäude, Rämistr. 101
http://www.ethz.ch/about/location/zentrum

Time

8:30 - 17:00 both days

Prerequisites

In order to make the course as productive as possible, all participants should come with an application to analyze, preferably from their research area. One or more batch scripts and corresponding input datasets for running short/small and longer/larger jobs on the Cray XE6   systems at CSCS should be prepared.  Participants are strongly encouraged to spend a day of preparation trying to instrument the application in advance of the course (instructions will be provided).  During the hands-on sessions, participants will learn how to augment and refine the instrumentation, obtain useful insight into the performance of their codes, and evaluate suggested optimization opportunities.  Each participant should bring along a laptop for wireless access.

Maximum number of participants

25

Target audience

This course is specifically aimed at HP2C users.

Accommodation

Participants are kindly requested to make their own arrangements for accommodation.

Course Syllabus

Monday, Jan. 17, 2010

This will be a fast-paced tutorial on the Scalasca Toolset, given by Dr. Brian Wylie of the Forschungszentrum Juelich (FZJ). 

1) Brief tutorial on the Scalasca toolset (morning)

a) Background information, application examples
b) Overview of different methods of instrumenting applications
c) Three stages of profiling: instrumentation, analysis, examination
d) Overview of instrumention types for user-level routines
e) Techniques for semi-automatic (with POMP) and manual instrumentation of individual routines and loops
f) Generating event traces
g) Running the application; generation of performance data
h) Analyzing performance data with CUBE3, other viewers
i) Filtering of frequently called small subroutines
j) Analysis of MPI events
k) Observing MPI I/O bandwidth and volume
l) PAPI hardware counters
m) Support for OpenMP and OpenMP/MPI-hybrid codes
n) Estimation of tool intrusiveness
o) Hands-on training with sample codes (provided)

2) Hands-on analysis of own benchmark application (afternoon)

a) Review of instrumentation process, address any problems with individual applications
b) Determination of routines requiring further investigation, custom profiling if needed
c) Running on the Cray XE6, address any problems with individual application
d) Refinement of instrumentation to the users' needs, possible manual instrumentation of individual routines or loops
e) Optional: evaluation of support for OpenMP or OpenMP/MPI-hybrid mode
f) Optional: brief presentation of results of individual applications

References:
Scalasca User's Guide  www.fz-juelich.de/js/datapool/scalasca/UserGuide.pdf

Tuesday, January 18, 2011

This tutorial on the Tuning and Analysis Utilities (TAU) toolset will be given by Dr. Sameer Shende, president of ParaTools Inc.

1) Brief tutorial on the TAU toolset (morning)

a) Background information, application examples
b) Overview of different methods of instrumenting applications
c) Custom profiling
d) Techniques for manual instrumentation of individual routines
e) Generating event traces
f) Running the application; generation of performance data
g) Analyzing performance data with ParaProf, PerfExplorer
h) Throttling effect of frequently called small subroutines
i) Observing I/O bandwidth and volume
j) PAPI hardware counters
k) Support for OpenMP and OpenMP/MPI-hybrid codes
l) Estimation of tool intrusiveness
m) TAU's interface to Scalasca
n) Hands-on training with sample codes (provided)

2) Hands-on analysis of own benchmark application (afternoon)

a) Review of instrumentation process, address any problems with individual applications
b) Determination of routines requiring further investigation, custom profiling if needed
c) Running on the Cray XE6, address any problems with individual application
d) Analysis of communication, input/output, scalability, Flop/s using ParaProf and/or PerfExplorer
e) Refinement of instrumentation to the users' needs, possible manual instrumentation of individual routines or loops
f) Optional: evaluation of support for OpenMP or OpenMP/MPI-hybrid mode
g) Optional: brief presentation of results of individual applications
             
References:
TAU User Guide  tau.uoregon.edu/tau-usersguide.pdf


Back to listing