Analysis of large parallel programs

Solving big problems (weather forecasting, simulation of astronomical events, ...) often requires parallel systems for high-performance computing. The runtime of such programs often takes a lot of time (several days, or even several weeks). During this time, there is no possibility to look inside of the calculation without a special mechanism. Users often want to assist and guide these long running computations using their domain knowledge.
The goal of our project [4] is to be able of on-line monitoring & steering of any arbitrary active program in memory. The main advantage of our system is, that there is no need to modify the source codes and to recompile the program. Therefore, the user needs not restart the program. Since large amount of data is provided by parallel programs, it seems natural to use graphical visualization.
Due to a big variety of existing visualization tools, we had to decide for one for us suitable tool. We voted for the Data Explorer from IBM to display provisional results of the calculation process, because of several advantages:

• it is an open-source software
• it has a client-server architecture
• it supports a lot of platforms
• it takes full advantage of multi-processor systems

One part of this project is the visualization of programs themselves, and this is the topic of this paper. The user should be able to feel the program, which is lying in memory. Different representation of programs will be described in one of the next sessions.