Scalasca  (Scalasca 2.0, revision 9811)
Scalable Performance Analysis of Large-Scale Applications
Introduction

Supercomputing is a key technology of modern science and engineering, indispensable to solve critical problems of high complexity. However, since the number of cores on modern supercomputers is increasing from generation to generation, HPC applications are required to harness much higher degrees of parallelism to satisfy their growing demand for computing power. Therefore – as a prerequisite for the productive use of today's large-scale computing systems – the HPC community needs powerful and robust performance analysis tools that make the optimization of parallel applications both more effective and more efficient.

Jointly developed at the Jülich Supercomputing Centre and the German Research School for Simulation Sciences (Aachen), Scalasca is a performance analysis toolset that has been specifically designed for use on large-scale systems such as IBM Blue Gene and Cray XT, but also suitable for smaller HPC platforms. Scalasca supports an incremental performance analysis process for applications using MPI[2] and/or OpenMP[7] which integrates runtime summaries with in-depth studies of concurrent behavior via event tracing, adopting a strategy of successively refined measurement configurations[4]. A distinctive feature of Scalasca is the ability to identify wait states that occur, for example, as a result of unevenly distributed workloads. Especially when trying to scale communication intensive applications to large processor counts, such wait states can present severe challenges to achieving good performance. Compared to its predecessor KOJAK[10], Scalasca can detect such wait states even in very large configurations of processes using a novel parallel trace-analysis scheme[3].

In contrast to previous versions of Scalasca which used a custom measurement system and trace data format, the Scalasca 2.x release series is based on the community instrumentation and measurement infrastructure Score-P [6], which is jointly developed by a consortium of partners from Germany and the US (see[8] for details). This significantly improves interoperability with other performance analysis tool suites such as Vampir[5] and TAU[9] due to the usage of the two common data formats CUBE4 for profiles and the Open Trace Format 2 (OTF2)[1] for event trace data. Nevertheless, the Scalasca 2.x series provides a certain level of backward compatibility, that is, the trace analysis component of Scalasca v2 is still able to process trace files generated by Scalasca 1.x.


Scalasca    Copyright © 1998–2013 Forschungszentrum Jülich GmbH, Jülich Supercomputing Centre
Copyright © 2009–2013 German Research School for Simulation Sciences GmbH, Laboratory for Parallel Programming