Experimental Physics and
Industrial Control System

Eric Norum <[email protected]> · Fri, 24 Sep 2010 09:16:25 -0700

On Sep 24, 2010, at 8:52 AM, Andrew Johnson wrote:

> On Friday 24 September 2010 10:34:28 Eric Norum wrote:
>> I don't understand the issue.
>> We run multiple soft IOCs on the same linux box all the time.   They each
>> get a port assigned by the kernel.  caRepeater takes care of fanning out
>> the UDP requests for PV information. "It just works".
> 
> ... unless someone configures a client with EPICS_CA_AUTO_ADDR_LIST=NO and 
> EPICS_CA_ADDR_LIST pointing to the IP address of that linux box.  With that 
> configuration each individual CA name search UDP message gets routed by the IP 
> kernel to only one of the IOCs, and depending on the particular kernel that 
> might always be the same IOC (thus hiding the other IOCs from that client) or 
> it might round-robin between IOCs, making connection times longer.

Ah, right.

> 
> One solution that I don't think anyone has suggested yet would be to run a CA 
> name-server on each IOC box at the official port number, and have it know 
> about and forward searches to all the IOCs on that particular box, or use the 
> "dbl > file" approach on each IOC so the name-server knows all the PVs 
> directly.  This needs no special configuration on the part of the clients, 
> they can use broadcasts or directed UDP as they wish, and there's no single 
> point of failure for the control system as a whole (assuming the name servers 
> get started when each machine comes up).

Or run every IOC in it's own virtual machine?

-- 
Eric Norum
[email protected]

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System