Experimental Physics and
Industrial Control System

Ralph Lange <[email protected]> · Thu, 04 Jan 2007 16:06:24 +0100

Well, not really problems, but difficulties and ambiguities:
As far as I recall, in some implementations of UDP/IP, only one of the
processes that share a UDP port will get a UDP unicast. This will break
the Channel Access name resolution mechanism, which relies on CA servers
being able to receive the resolution requests. In most (maybe all
newer?) cases all the port-sharing processes will get the message, which
works. That's what you see on your Linux box (also true for Windows, I
heard).
If I remember correctly, there was a problem with CA beacons coming from
those multiple soft IOCs: Clients were interpreting the regular beacons
from multiple IOCs residing on the same IP as irregular beacons from one
server and continuously retransmitting all their unresolved channels. As
far as I know Jeff has fixed this by adding IP and port info to the
beacon packets, so that the clients can distinguish these cases.
I still prefer having different IP addresses for different IOCs on the
same host, as you can find out a lot easier which soft IOC a certain
channel (that an external CA client is connected to) resides on. Useful
when things are hanging and you want to know which of the soft IOCs has
to be rebooted.
I am not expert enough to tell what positive or negative side effects
multiple logical interfaces (with their separate stacks and system
resources) have - compared to running all traffic over one interface.
Somehow keeping things separate feels more adequate to me, but that
feeling may be wrong just as well.
Cheers,
Ralph
ps. Note that the addition of logical interfaces does not require any
additional hardware in your Linux box, it's just creating new network
interfaces that share the same ethernet card. Pure configuration.

Luedeke Andreas wrote:

Hi,

at the SLS we do run many IOCs on one Linux Host (Redhat 7.3, Redhat 9
or ScientificLinux 3).

We get the same error message, but we just ignore them:

until now we did not run into any problems due to this.

We just run "soft-iocs" on Linux without any hardware connections.
Ralph, where would you expect a problem from this shared UDP port?
Kind Regards
Andreas

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System