The Paradyn parallel performance measurement tool

Abstract

The Paradyn performance-measurement tool uses several novel technologies to scale to long-running programs and large systems and automate much of the search for performance bottlenecks. The tool is based on a dynamic notion of performance instrumentation and measurement. Application programs are placed into execution, and performance instrumentation is inserted into the running programs and modified during execution. The performance consultant module automatically directs instrumentation placement and associates bottlenecks with specific causes and specific program parts. Paradyn controls its instrumentation overhead by monitoring the cost of its data collection and limiting its instrumentation to a user-controllable threshold. Paradyn's instrumentation can be configured to accept new operating systems, hardware, and application-specific performance data. Paradyn also provides an open interface for performance visualization and a simple programming library to interface visualizations. Paradyn can gather and present performance data in terms of high-level parallel languages (such as data-parallel Fortran) and can measure programs on massively parallel computers, workstation clusters, and heterogeneous combinations of these systems. The article presents two examples that illustrate Paradyn's use in finding performance problems in real applications.