Paper Details

%T The Investigation of Communications Patterns in Occam Programs
%A Rosemary Candlin, Qiangyi Luo, Neil Skilling
%E J. Wexler
%B OUG\-11: Developing Transputer Applications
%X The performance of a concurrent computation running on a
   multiprocessor system may depend critically on the way in
   which the program is decomposed and placed on the machine.
   In order to exploit the potential of parallel processors, it
   is necessary to balance the advantage of spreading the
   computational load as thinly as possible over the
   processors, with the disadvantage that increased
   communication delays may slow down the computation. In
   general, there is no satisfactory theoretical model of the
   complex interaction between the amount of computation
   carried out by the individual processes, their frequency of
   communication and the topology of the underlying machine.
   For many programs, it is not easy to see in advance how
   computation will interact with communication, and placement
   strategies which depend only on a static analysis of the
   program structure may not be sufficient. The work described
   here is an attempt to provide useful tools for the occam
   programmer which can be used to investigate communications
   patterns, and to explore different configurations rapidly.We
   think that this approach will be particularly valuable for
   programs which can be decomposed in a natural way into a
   fairly large number of top\-level occam processes, so that
   the preliminary parallelization arises out of the nature of
   the application, and the main problem is to place these
   processes on a smaller number of physical processors. This
   is often the case for programs which model real\-time
   systems, and we have taken as an example an application from
   chemical engineering. In programs like this, there is
   natural concurrency in the real world which can be easily
   represented in terms of occam processes. At the moment, we
   do not attempt to extract parallelism automatically, or
   handle shared data, though there are a number of systems
   that have tackled these problems (see, for example [1] and
   [2]). Our main aim at this stage is to provide a programmer
   with profiling tools, and see to what extent they can help
   to produce an efficient implementation of the program.

If you have any comments on this database, including inaccuracies, requests to remove or add information, or suggestions for improvement, the WoTUG web team are happy to hear of them. We will do our best to resolve problems to everyone's satisfaction.

Copyright for the papers presented in this database normally resides with the authors; please contact them directly for more information. Addresses are normally presented in the full paper.