Experimental Physics and
Industrial Control System

Michael Davidsaver via Tech-talk <[email protected]> · Wed, 22 Jan 2020 12:46:00 -0800

On 1/22/20 12:10 PM, Kasemir, Kay via Tech-talk wrote:
> PV Access _is_ using multicasts internally, basically as a replacement of the ca repeater.
> When multiple servers and/or clients run on the same host, they pass the UDP messages which would by default only reach one of them to each other via a multicast.
> For that, it makes sense to have a TTL of 1.
> Would it hurt to send those with a larger TTL, when that enables using multicasts beyond localhost?

No.  Simply because it wouldn't do anything.  The socket involved is joined/bound
to the loopback interface.  So this traffic will (and should) never appear on a
physical wire.

What is needed is to join the individual broadcast sending/receiving sockets to
an appropriate mcast group.  And provide a way to set the TTL.  This should not
a major change.  The work for CA was a smaller follow on to a larger rework
of the RSRV code to make $EPICS_CAS_INTF_ADDR_LIST work as expected.

https://code.launchpad.net/~epics-core/epics-base/rsrvbindiface/+merge/282249

> -Kay
> 
> ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> *From:* Tech-talk <[email protected]> on behalf of Michael Davidsaver via Tech-talk <[email protected]>
> *Sent:* Wednesday, January 22, 2020 2:24 PM
> *To:* Denise Finstrom <[email protected]>
> *Cc:* Jim Smedinghoff <[email protected]>; [email protected] <[email protected]>
> *Subject:* [EXTERNAL] Re: EPICS_CA_MCAST_TTL equivalent for pvAccess
>  
> On 1/22/20 8:18 AM, Denise Finstrom via Tech-talk wrote:
>> 
>> Hello,
>> 
>> Is there an equivalent to EPICS_CA_MCAST_TTL for the pvAccess protocol?
> 
> There isn't one.  PVA doesn't (yet) support search by mcast.
> 
> Were you able to make this work with CA?  What does your network topology look
> like?  (how many hops)  Did you encounter any problems?
> 
> Adding this was a project of mine a few years ago.  I've actually never been
> certain that it would work at scale since I don't have access to a large
> network with mcast routing enabled for testing.  Nor have I had any feedback
> from users until now.  So adding this feature to the PVA protocol code hasn't
> been a priority for me.
> 
> 
>> We are attempting to use multicast for PV search and haven't found a way to change TTL values when using pvAccess. When we set EPICS_PVA_ADDR_LIST to a multicast address, the multicast goes out with a TTL of one and doesn't get routed to other subnets. We are successfully using multicast for PV search with the CA protocol and EPICS_CA_MCAST_TTL. We are looking to do the same with pvAccess, but the EPICS_PVA_MCAST_TTL environment variable doesn't seem to exist. We are using base-7.0.3.1. Any advice would be appreciated.
>> 
>> Thanks,
>> 
>> Denise
> 

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System