Deadlock free routingMy first reaction would be to refer back to Peter's High Level Paradigms paper of WoTUG 19 (?) and to the various forms in which Jeremy Martin's work has been published.We have a discussion about deadlock-free routing.
Sirs 26 June 1997
In order to guarantee deadlock-free routing:
Q1. What is the relationship between how many buffers we need
and the blocking characteristics of application-code?
Briefly: Client-Server networks cannot deadlock if they are free of cycles; IO-Par and IO-Seq networks are also deadlock free. (Other networks may or may not be deadlock free.)
One of the exciting corollories of these theoretical results is the rather counter-intuitive discovery that less buffering is needed. This is interesting because most informal approaches (such as Schlaer-Mellor - admitedly an analysis technique but one involving communicating objects) rely on large buffers to operate correctly. Often what happens when buffers run out is ill-defined.
It is difficult to have really large buffers in occam because there is no support for virtual memory. It's just as well that deadlock freedom can be had with no buffering or just a little buffering of fixed-size - this takes less memory and probably executes more efficiently too. :-)
Right. Time to stop thinking aloud and leave the subject to someone who really knows what they're talking about!
Regards,
Rick
--
Richard Beton BSc CPhys MInstP
Roke Manor Research Limited
--------- Standard Disclaimer about my own views etc etc --------
Welsh Highland Railway: http://www.whr.co.uk/WHR/WHR.html