Experimental Physics and
Industrial Control System

Ben Franksen <[email protected]> · Sun, 4 Nov 2007 21:23:16 +0100

On Freitag, 2. November 2007, Eric Norum wrote:
> On Nov 2, 2007, at 8:30 AM, Ned Arnold wrote:
> > You may be able to eliminate the fanout records by careful design
> > of forward links and/or input link flags (PP, CP, etc).
> >
> > For example, if the input record's forward link field referenced
> > the calc record and the calc record's forward link field referenced
> > the calcout record, all would process each time the input record
> > processed ... without a fanout.
>
> Don't take this technique to extremes, though.  I recall helping
> Steve Shoaf diagnose problems with his database for a video switching
> system.  The database contained 128 records FLNK'd together.  This
> caused very weird IOC faults that we finally realized were caused by
> stack overflow of the task processing the records -- 128 recursive
> calls to the record processing routine forced the task stack pointer
> past the end of the stack space.

Indeed, in one application (a feed-forward from insertion device positions 
to quadrupole currents) that connects about 30 inputs and 40 outputs and 
does some non-trivial processing in-between, I found the system to behave a 
lot more reliable and predictable if all records get periodically scanned, 
compared to the use of .CP links, FLNKs etc. Order of processing between 
the stages is imposed via setting the PHAS field (I use 4 different 
values).

Cheers
Ben

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System