Re: evaluation of OMG DDS as a controls protocol for EPICS

[Claude Saunders <saunders@aps.anl.gov>]

I've been watching DDS from the sidelines as well, and only recently 
decided to take the plunge. It really has not achieved widespread 
support, and I'm not sure if that's because of limitations like those I 
discovered, or because it didn't reallly follow what I would consider a 
normal progression towards being a standard. It started out as 
proprietary, then became an odd API-only "standard" through OMG for some 
time. Clearly not interoperable. Only later was a wire protocol 
released. Even so, there are so many options and extensible areas of the 
specification that it seems unlikely initial implementations will play 
well together. The DDS spec is really quite recent (2004), so if CORBA 
is any guide, we should see this mature in about 10 years :)

I only looked at a few of the QoS capabilities implemented by ORTE, but 
the RTPS specification is an interesting read in this area, and can give 
lots of ideas. For example, they consider timestamping of messages a QoS 
option. Also, you can have multiple publishers of the same message type 
on the same topic, each with a relative "strength". Subscribers can 
indicate that they want a message from strength "2" publishers first, 
unless some time window passes without a message, in which case the 
subscriber will accept a message from strength "1" publishers. So in 
this manner one could implement a form of failover. The specification 
also has something called "Content-Filtered Topics", which is much like 
the "beam flavor subscriptions" that Jeff has been talking about for CA. 
The filtering can be implemented on the subscriber side, or on the 
publisher side (preferred, but more complex). There is also something 
called "Coherent Sets", which seems to be similar to the channel access 
"sync groups". At any rate, it's interesting to find the parallels. I 
think DDS makes many of these concepts more explicit.

Just out of curiosity, what other RTI S/W have you used? Other than 
NDDS, I'm only aware of their SkyBoard product (which is very odd and 
interesting at the same time).

  regards, Claude

Lawrence T. Hoff wrote:
>    Thanks for sharing, Claude!
>
>    I have been watching DDS "from the sidelines" for some time now.
>
>    Ignoring the technical details, my chief reservation is that DDS 
> does not
> seem to be achieving widespread support, despite active marketing by RTI.
> Disclaimer: I am actually a very satisfied user of other RTI S/W, but 
> trying
> to use a communication standard that is not widely adopted gives me 
> visions
> of trying to make hotel reservations in Esperanto. :-)
>
>    On the plus side, DDS does provide "connection management" (built
> on top of UDP/IP), and (more importantly, perhaps) includes some level
> of support for user-defined Quality of Service (QoS) parameters when
> establishing connections.
>
>    Did you learn anything about DDS's QoS capabilities that could
> (theoretically) be applied to CA?
>
> -- Larry