Rainer Koppler

Parallel Program Visualization with MUCH

The use of visualization in parallel program development is manifold. It is applied from data and control flow over debugging, performance analysis and performance prediction until data distribution on distributed memory architectures. Most of these visualizations do not care about the physical topology of the underlying hardware although this can be of importance in the fields of performance analysis or error debugging.

Geometry-Aided Rectilinear Partitioning of Unstructured Meshes

This paper introduces an infrastructure for parallel mesh computations running on distributed-memory computers. The infrastructure consists of the mesh partitioning algorithm GARP and the domain-specific communication library GRAPHlib. Unlike existing algorithms, GARP exploits geometrical properties of the mesh shape in order to produce shape-adequate rectilinear partitions. The structure of such partitions is exploited by GRAPHlib using an optimized message ordering strategy. We describe the concepts behind GARP and GRAPHlib and show that for meshes with particular shapes our infrastructure provides better utilization of the parallel computer than solutions using existing partitioning algorithms and communication libraries.

Visualization of distributed data structures for High Performance Fortran-like languages

This article motivates the usage of graphics and visualization for efficient utilization of High Performance Fortrans (HPFs) data distribution facilities. It proposes a graphical toolkit consisting of exploratory and estimation tools which allow the programmer to navigate through complex distributions and to obtain graphical ratings with respect to load distribution and communication. The toolkit has been implemented in a mapping design and visualization tool which is coupled with a compilation system for the HPF predecessor Vienna Fortran. Since this language covers a superset of HPFs facilities, the tool may also be used for visualization of HPF data structures.

Exdasy - A User-Friendly and Extendable Data Distribution System

This paper introduces Exdasy, a user-friendly and extendable software tool for partitioning unstructured meshes anti mapping mesh partitions to parallel computers. Exdasy was designed to meet the increasing demands to todays data distribution systems, which are posed by the variety of mesh computations, the ongoing development of distribution algorithms, and rapid changes in parallel hardware technology. For this, Exdasy offers third-party state-of-the-art distribution algorithms augmented with graphical user interfaces and powerful graphical evaluation displays. Evaluation of distributions is based on various quality metrics and static machine parameters. Exdasy provides a modular architecture by means of replaceable distribution algorithms, machine models and evaluation facilities. Hereby it is attractive to both users and developers.

SPiDER—An advanced symbolic debugger for Fortran 90/HPF programs

Debuggers play an important role in developing parallel applications. They are used to control the state of many processes, to present distributed information in a concise and clear way, to observe the execution behavior, and to detect and locate programming errors. More sophisticated debugging systems also try to improve understanding of global execution behavior and intricate details of a program. In this paper we describe the design and implementation of SPiDER, which is an interactive source‐level debugging system for both regular and irregular High‐Performance Fortran (HPF) programs. SPiDER combines a base debugging system for message‐passing programs with a high‐level debugger that interfaces with an HPF compiler. SPiDER, in addition to conventional debugging functionality, allows a single process of a parallel program to be expected or the entire program to be examined from a global point of view. A sophisticated visualization system has been developed and included in SPiDER to visualize data distributions, data‐to‐processor mapping relationships, and array values. SPiDER enables a programmer to dynamically change data distributions as well as array values. For arrays whose distribution can change during program execution, an animated replay displays the distribution sequence together with the associated source code location. Array values can be stored at individual execution points and compared against each other to examine execution behavior (e.g. convergence behavior of a numerical algorithm). Finally, SPiDER also offers limited support to evaluate the performance of parallel programs through a graphical load diagram. SPiDER has been fully implemented and is currently being used for the development of various real‐world applications. Several experiments are presented that demonstrate the usefulness of SPiDER. Copyright

Graphical support for data distribution in SPMD parallelization environments

The paper introduces a programming environment for HPF like languages with emphasis on graphical support for data distribution. A novel component of this environment is a mapping design and visualization tool. The tool provides visualization of HPF array objects such as data arrays and logical processor arrays and creates a number of diagrams based on information that is gathered from other components of the environment such as the compiler or a debugger. The diagrams relate to crucial issues such as load distribution and communication. Furthermore we show how our environment facilitates seamless integration of additional components.

Debugging MPI Programs with Array Visualization

Debugging message passing programs is accepted as one of the major difficulties of parallel software engineering. Besides various problems with communication and synchronization of concurrently executing processes, one of the big obstacles is the amount of data that is processed in parallel applications. Yet, inspecting these data is a basic necessity to verify the correctness of a program. Therefore the MAD environment includes an array visualization component, which displays arbitrary arrays distributed on parallel processes in MPI programs. Although the current implementation restricts arrays to HPF-like distributions, the usefulness of this first prototype already indicates how vital such a visualization feature can be for parallel program debugging.

Parallelization of Unstructured Mesh Computations Using Data Structure Formalization

This paper introduces a concept for semi-automatic parallelization of unstructured mesh computations called data structure formalization. Unlike existing concepts it does not expect knowledge about parallelism but just enough knowledge about the application semantics such that a formal description of the data structure implementation can be given. The parallelization tool Parlamat uses this description for deduction of additional information about arrays and loops such that more efficient parallelization can be achieved than with general tools. We give a brief overview of our data structure modelling language and first experiences with Parlamat’s capabilities by means of the translation of some real-size applications from Fortran 77 to HPF.

Advanced visualization and data distribution steering in an HPF parallelization environment.

On distributed-memory systems, the quality of the data distribution has a crucial impact on the eeciency of the computation. Sophisticated visualization of large in-core and out-of-core data, and steering capabilities of the debugging system sig-niicantly reduce program development cycle, especially for irregular applications. In this paper we present an advanced system for visualization and steering of data distributions based on the Graphical Data Distribution Tool (GDDT) and the advanced HPF symbolic debugger SPiDER. The development focused on eecient support for regular and irregular data distributions and brings several signiicant contributions: (1) on-line visualization of distributions and values of large in-core and out-of-core data structures, (2) quality control of data distributions using (re)distribution animation, and (3) dynamic data redistribution in parallel programs.