Visualizing performance debugging

Abstract

This article discusses visualizing performance of parallel and sequential computations using the Programming and Instrumentation Environment (PIE) as an example. First, the importance of being able to visualize the performance of a computation is demonstrated. Secondly, an easy guide to one way, the PIE way, of his visualization process is presented. This is followed by examples of concrete uses of the environment. Finally, some of the pressing issues concerning measuring performance are discussed.Three issues are stressed in this article:1. Visualization of program performance is necessary for fast, correct performance debugging.2. PIE is a powerful and useful implementation for supporting visualization. It is also a dramatic teaching assistant in that it visually shows actions like forking, scheduling, etc.3. PIE provides complete support for visualization. The system is aware of issues like program replay, suitably presenting performance data, compensation for perturbation by the measurement facility and support for multiple architectures and languages .

PProto: An environment for prototyping parallel programs

“…ParaScope assists in the formulation, implementation, automatic parallelization, computation scheduling, and debugging of parallel Fortran programs [5]. The Parallel Programming and Instrumentation Environment (PIE) focuses on visualization of concurrent program performance for debugging and supports analysis, verification, and validation of a computation's performance [6]. In Faust, the emphasis is on an integrated environment for scientific applications that includes database, data and control analysis, restructuring, program maintenance, and performance analysis components [7].…”

LBF: a performance metric for program reorganization

“…Profile metrics [1,6,15,22] associate a value with each component of a distributed or parallel application (frequently procedures), and are presented as sorted tables. Visualizations [8,13,14,18,23] explain application performance using pictures. Search tools [10,17,21] help users to manage performance data information overload by treating the problem of finding a performance bottleneck as a search problem.…”

Visualization: An Aid to Design and Understand Neural Networks in a Parallel Environment

“…The motivation behind such transformation is to get a better insight of the computational process. Some examples of applications in this category are computational fluid dynamics [3), finite element analysis [l], parallel program development and debugging [5].…”