Debugging parallel programs in parallel

• Authors: Robert Rubin;Larry Rudolph;Dror Zernik
• Venue: PADD '88 Proceedings of the 1988 ACM SIGPLAN and SIGOPS workshop on Parallel and distributed debugging
• Year: 1988

There often exists a large gap between the user's conceptual picture of a parallel machine and the real machine, and between the user's expectations for program execution and the real process of execution. A monitoring and debugging machine can be used to bridge this gap. The Makbilan machine described here creates and maintains in real time a picture of the system that displays different layers of the machine. It can display hardware performance (such as bus accesses, memory references), operating-system abstractions (such as dynamic process tree or resource allocation), and special-purpose animations of the program (designed by the programmer in a constraint-based graphical environment).This paper introduces the design of a non-intrusive monitor with a constraint-based graphic display machine, and is event-driven. The Makbilan machine on which the parallel programs run is under construction, and the monitor is thus designed and simulated in concordance with hardware specifications. The contributions of this paper are: The use of interesting events to integrate a monitor, a debugger and an animation tool for parallel algorithms.A non-intrusive architecture for monitoring executing programs.A flexible animation tool and a set of animations that can be used to understand the parallel structures of parallel algorithms and their execution on a parallel machine.