Scalasca can be used with parallel codes of any size, from multicore machines to the fastest supercomputers in the world. It supports MPI and/or OpenMP libraries. It supports Blue Gene, Cray XT and Linux clusters.

Scalasca combines runtime summaries suitable to obtain a performance overview with in-depth studies of concurrent behavior via event tracing. The traces are analyzed to identify wait states that occur, for example, as a result of unevenly distributed workloads.

The new version of Scalasca is based on the community instrumentation and measurement infrastructure Score-P. This improves interoperability with other performance analysis tools, by the use of the Open Trace Format 2 for event trace data, and CUBE4 for application profiles.

CUBE is the interactive analysis report explorer GUI for Scalasca. Its color-coded scheme makes observation of timing phenomena straightforward.

The talk will introduce the main components of Scalasca to the audience.

Topics will cover:

• Examples of delays caused by imbalances
• Propagation effects of such imbalances
• How to spot these events in GUI

After the audience understands why the imbalances exist, a quick how-to follows, showing

• Instrumenting the code
• Measurement
• Analysis
• Evaluation