Jste zde

Practical Performance Analysis of Parallel Applications

Termín: 
Út, 27.05.2014 9:30 - St, 28.05.2014 17:00
Místo: 
VŠB - Technical University Ostrava, building NA, auditorium NA4
Lektor: 
Brian Wylie (Jülich Supercomputing Centre ), Bert Wesarg, Thomas William (TU Dresden)
Úroveň: 
intermediate
Jazyk: 
English

Annotation

Applications need to be effectively parallelised, using techniques such as MPI message-passing and OpenMP multithreading, to fully exploit the capabilities of modern systems for high-performance computing.The course introduces the open-source community-developed Score-P instrumentation and measurement infrastructure, and the Scalasca and Vampir tools using it, to provide a practical basis for portable performance analysis of parallel applications.

The course will be delivered as a series of presentations with associated hands-on practical exercises using the Anselm computer cluster.  It starts with basic application instrumentation and measurement to generate execution profiles, then improves measurement quality via customization capabilities, and progresses to interactive and automated analyses of execution traces.

While analysis of provided example codes will be used to guide the class through the relevant steps and familiarise with usage of the tools, coaching will also be available to assist participants to analyse their own parallel application codes and may suggest opportunities for improving their execution performance and scalability.

Purpose of the course (benefits for the attendees)

On completion participants should be familiar with common performance analysis techniques and how they can be employed in practice (on a range of HPC systems).  Those who prepared their own application test cases will have been coached in the tuning of their measurement and analysis, and provided optimization suggestions.

Schedule

Tuesday, May 27, 2014
09:30-10:00registration
10:00-11:30Introduction to parallel performance engineering
 Instrumentation and measurement of parallel applications with Score-P
11:30-12:30time for lunch
12:30-14:00Profile analysis report exploration
 Configuration and customisation of Score-P measurements
14:00-14:30coffee break
14:30-16:00Coaching to profile your own parallel application(s) on Anselm
16:00-16:30coffee break
16:30-18:00Parallel execution performance analysis diagnosis and tuning
 
Wednesday, May 28, 2014
09:00-10:30Automated trace analysis with Scalasca            
10:30-11:00
coffee break
11:00-12:30Interactive trace analysis with Vampir
12:30-13:30time for lunch
13:30-15:00Coaching to trace your own parallel application(s) on Anselm
15:00-15:30coffee break
15:30-17:00  Parallel execution performance analysis diagnosis and tuning

Prerequisities

Familiarity with MPI or OpenMP programming in Fortran, C or C++, and running parallel applications on Linux clusters. Ideally participants will have their own parallel application code which
builds and runs successfully on the Anselm cluster.  Participants need to have their own notebook computer configured for access to Anselm and running X11-based graphical tools.

Registration

Obligatory registration - registration form here; deadline May 21, 2014 or exhausted course capacity.

Capacity

35 attendees

Practicalities

  • The venue building is marked as "NA" on the campus map. See a special page on transport (in Czech) how to get there.
  • Do not forget your notebook.
  • Anselm training accounts will be distributed at the registration. Ask for an account in advance, if you need to prepare your application before the course.
  • Anselm cluster documentation is available at http://support.it4i.cz/docs.