Experimental Physics and
Industrial Control System

[email protected] (Ned Arnold) · Fri, 3 May 1996 13:59:37 -0500

> Dear all,
> 
>   Thanks to Steven, Jeff and Andrew for their replies. I guess I _knew_
> that if I mentioned the reason for the question then I would get
> comments on the proposed application too! Thanks for those, and yes, of
> course one would have to be very careful at the switchover point.
> 
> Andrew said:
> > If you know that the Unix CA server will always be running, you could 
> > always have the Unix server reflect PVs from the IOC into a different 
> > namespace, and have all your clients attach to the Unix names and never to 
> > the original IOC names.  This way the Unix server knows exactly when it 
> > needs to be a simulator (because it gets connect and disconnect messages 
> > from the IOC), and the clients couldn't care less if an IOC gets turned 
> > off because they're never talking directly to it. There will be a 
> > performance hit here of course, but for slow records it shouldn't matter.
> 
>   This is close to the existing non-EPICS system that we are trying to
> emulate.  Things are a lot easier if all control and status flows
> through a single point in both simulate and non-simulate modes because
> the simulate mode can be a single global flag at that single point. I
> was wondering whether there was a solution which still allowed clients
> to connect directly to the online system.

Actually, all the pieces to <at least one> solution to this problem already
exist !  If you replace the term "UNIX server" with the words "vxWorks server"
your almost there. 

One could dedicate an IOC to be the "server". Lots of "soft records" would
be used to interact with the process, each of which has channel access
links to the REAL PV's (on other IOC's). It also has additional soft records
for the simulation input and output. Using the SIMULATION MODE in the 
standard records, a simple change of one process variable ("SIM_MODE_BI") would
cause the records to fetch their input (and put their outputs) from/to the 
simulation soft records instead of the REAL records. One could even 
"automatically" switch to the simulated PV's when the CA_LINK to the REAL
PV went INVALID (this would take a CALC record). You could even notify the
operator that simulation mode was entered via another existing tool, the
Alarm Handler.

This solution might get a little ugly for a BIG application, but it is the
only one possible right now because the vxWorks environment is the only place
we have "client" AND "server" capabilities, despite promises to the contrary.

		Ned

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System