Array visualization for parallel program debugging

Debugging is one of the most difficult activities of the parallel software life-cycle. This fact can be attributed to malign errors like deadlocks and race conditions, as well ar; huge amounts of data that are distributed on concurrently executing processes. To detect program errors, debuggers must support users in extracting, inspecting, and manipulating variable values. While this is usually rather trivial for basic data types, evaluating derived types like arrays or complex data structures may impose severe problems. This issue is addressed in the MAD environment with a visualization component for displaying a program's execution as an event graph, and an array viewer showing distributed arrays as heat diagrams. This powerful debugging feature is described with examples of actual programming errors as they occurred in a case study.