University of Houston

The Abstract Data and Communication Library (ADCL) helps meet the performance requirements of parallel applications by optimizing the communication operations for large scale simulations at runtime, adapting to the current hardware and software environment. The ease with which ADCL can be incorporated into an application is one of its main advantages, yet the output produced is not readily understandable and useful for the user. Such a task can be accomplished by a visualization and analysis tool. Even though research and development in the area of visualization and performance analysis of parallel applications has resulted in the emergence of several tools, they are suitable for post mortem analysis of application using MPI generated trace files and do not meet the requirements of a visualization tool for ADCL.

The contribution of this thesis is the development of the recommendation system, a visualization and analysis tool tailored for use with ADCL. The recommendation system provides hints to the user about which implementation is optimal for certain communication patterns on a particular platform. Additionally, it provides visual representation of the data generated by the ADCL which makes it easier for users to understand and analyse the results. The recommendation system operates in two modes, one for representing and analysing live data received from MPI processes, another for reading and parsing through application generated output files to extract relevant information to facilitate visualization and analysis.