some old computer science

[Larry Dickson <tjoccam@xxxxxxxxxxx>]

Hello Peter and all,

I’m putting the finishing touches on my Kickstarter-promised book on “crawl-space computing” (a popularization of Wide computing, i.e. CSP and all its children), and ran across a very relevant reference: Welch, Justo, and Willcock, “High-Level Paradigms for Deadlock-Free High-Performance Systems,” Transputer Applications and Systems '93. The client-server part of it is pretty clear, but I have a couple of questions relating to the I/O-PAR and I/O-SEQ part:

(1a) You say “I/O-SEQ is usually applied in systems where the component interaction is not symmetric (e.g. logic circuits and control laws)”. Are there examples floating around of such an application?

(1b) The I/O-SPnet restriction “no closed loop or path from external input to external output consisting only of I/O-SEQ normal forms” sounds like a killer, especially the no path part - seems to exclude FIFOs - so how is it clear in (1a) that we can use I/O-SPnets for most applications?

(2) How do the two special cases(client/server and I/O-PAR-I/O-SEQ) combine to prove deadlock freedom for the hybrid? If you draw a line around all the ropes in Figure 17, you don’t get a pure client/server, because of the two channels labeled p and connecting with user interface.

The paper says “maintain a separate view of the different types of network as intact sub-systems, even though the same processes may be part" of both. I suspect I’m missing something obvious, but it sounds like interference would be possible.

Any help would be appreciated!

Larry