Experimental Physics and
Industrial Control System

"Kasemir, Kay" <[email protected]> · Wed, 11 Feb 2015 14:06:29 +0000

Hi:

I've often seen problems when trying to run multiple IOCs on a single non-networked host.
This is unlikely to happen in a production environment, but you might want to do this for a live demonstration on a laptop, or for development.

When the host is connected to a network, there's no problem: Add the broadcast address of that host to the EPICS_CA_AUTO_ADDR_LIST, and clients can connect. More on that is on https://wiki-ext.aps.anl.gov/epics/index.php/How_to_Make_Channel_Access_Reach_Multiple_Soft_IOCs_on_a_Linux_Host

The situation is different when the host is not connected to a real network, so all you have is "localhost" and no broadcast address.
As Ralph mentions, sometimes it works and sometimes it doesn't.
In addition, there seems to be some remaining difference between the C implementation of the channel access client library and the pure Java CAJ implementation. The only solution that always works: Somehow create a dummy network. Connect the host to a hub that's otherwise not connected to anything, just to get its ethernet light to come on, assign some IP address. Or enable the wireless network in 'peer' mode, again not really connected to anything, but you now have an IP and thus broadcast address that channel access can use on that host.

Thanks,
Kay

On Feb 11, 2015, at 8:42 AM, Ralph Lange <[email protected]> wrote:
> On February 11, 2015 2:02:15 PM GMT+01:00, Dirk Zimoch <[email protected]> wrote:
>> Since you have EPICS_CA_AUTO_ADDR_LIST=NO and 
>> EPICS_CA_ADDR_LIST="100.100.10.255 localhost", I suspect that you run 
>> the IOCs and caget on the same host and 100.100.10.255 is NOT the 
>> broadcast address of that host. Because in that case caget finds the 
>> channels at "localhost" and that is a unicast address with the known 
>> limitations.
> 
> I have seen 127.255.255.255 (i.e. the broadcast address of localhost) working. But I also have seen it not working, and I have no insight what exactly is needed in the network stack for it to work.
> Worth a try, in any case.
> 
> ~Ralph

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System