On 09/12/2018 03:34 PM, Stuart Stevenson wrote:
What is wrong with NML? If it doesn't have some needed process is it not
able to be extended?
The problem, AS I UNDERSTAND IT, is that NML treats the
entire real time state of the system as one atomic block.
Works fine on any system with one memory. So, multi-core
systems are fine. But, if you want to extend something that
uses NML communication across the network, it has to send
20+K of real time state across the network at the servo
rate. That IS a problem. NML can ONLY send the WHOLE real
time state. The feature of zmq (apparently they have now
converged on one way to spell it) is that it can send ONLY
the part that a requester asks for. So, if the remote thing
only needs a few variables out of the whole real time state,
ONLY those variables will be sent.
This is my understanding of the major feature of zmq that
they want to use. There MAY be a talk by one of the MK
developers about this that is more authoritative and goes
into vastly more detail. We had at least 2 MK meetings at
Tormach a few years ago, and there were some talks by
Michael Haberler and Alex Rossler, among others, that went
WAY over my head. I think some of these talks should be online.
Jon
_______________________________________________
Emc-developers mailing list
Emc-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-developers
