Difference between revisions of "Scalasca"

From SNIC Documentation
Jump to: navigation, search
(Overview)
Line 8: Line 8:
 
Scalasca is a performance analysis tool specially geared towards investigating parallel programs.  It can also be deployed to profile serial applications.
 
Scalasca is a performance analysis tool specially geared towards investigating parallel programs.  It can also be deployed to profile serial applications.
  
Scalasca can analyse programs written in [[Fortran]], [[C]] and [[C++]] which are parallelised in [[MPI]] and/or [[OpenMP]].
+
Scalasca can analyse programs written in [[Fortran]], [[C]] and [[C++]], which are parallelised in [[MPI]] and/or [[OpenMP]].  From the output one can learn where a parallel program is spending time.  The tool also highlights typical problems associated with parallel programming, such as load imbalance issues in the code.  If the hardware performance counter tool [[PAPI]] is installed on a system, this can be integrated into Scalasca.
 +
 
 +
Scalasca comes with a GUI to visualise the results, which aims to conveniently display the performance issues of an application, even if a large number of tasks and threads are deployed.  The results can also be analysed with a number of tools provided by other groups, such as e.g.: [[Paraver]].
  
 
== Availability ==
 
== Availability ==

Revision as of 15:05, 20 September 2012


Overview

Scalasca is a performance analysis tool specially geared towards investigating parallel programs. It can also be deployed to profile serial applications.

Scalasca can analyse programs written in Fortran, C and C++, which are parallelised in MPI and/or OpenMP. From the output one can learn where a parallel program is spending time. The tool also highlights typical problems associated with parallel programming, such as load imbalance issues in the code. If the hardware performance counter tool PAPI is installed on a system, this can be integrated into Scalasca.

Scalasca comes with a GUI to visualise the results, which aims to conveniently display the performance issues of an application, even if a large number of tasks and threads are deployed. The results can also be analysed with a number of tools provided by other groups, such as e.g.: Paraver.

Availability

ResourceCentreDescription
AlarikLUNARCthroughput cluster resource of 40 TFLOPS
AuroraLUNARCthroughput/general purpose cluster resource
PlatonLUNARCthroughput cluster resource of 26 TFLOPS
TriolithNSCCapability cluster with 338 TFLOPS peak and 1:2 Infiniband fat-tree

License

License: Free.

Free but copyright

Experts

These experts have registered specific competence on this subject:

  FieldAE FTEGeneral activities
Joachim Hein (LUNARC)LUNARCParallel programming
Performance optimisation
85Parallel programming support
Performance optimisation
HPC training

Links