Re: Asyn-based TCP/VXI-11 interface to many devices

[Rod Nussbaumer <bomr@triumf.ca>]

Hi Eric.

Which protocol gets used will be partly a function of which PS's are 
ultimately chosen. So far I haven't used VXI-11 on devices that support 
it, as I haven't actually seen any particular advantage over a raw TCP 
stream. If I've missed something, I could be persuaded to switch.

What significance do you put on TCP low-level connection maintenance at 
the IOC level. I've always treated that as a black-box that applications 
shouldn't need to look into. StreamDevice seems to do a fair job of 
handling timeouts in my experience so far. I am still a bit skeptical 
about the reliability of reconnecting to broken connections, and this is 
also going to be the subject of some exploration.
I have seen connections fail to re-connect in the past, and it's 
difficult assess whether the problem originates at the peer host, the 
OS, asyn,....

As you haven't noted any deficiencies, I assume that your interface to 
55 supplies are working properly?

Thanks.

   ---   rod.




On 05/10/2013 10:14 AM, Eric Norum wrote:
> We have 55 Caen A36xx supplies connected to a single Linux IOC process and 24 supplies to another.  Communication is over a raw TCP/IP stream using ASYN plus a small custom driver since there are some features of the command set that don't fit well with StreamDevice.  At some point I plan to try UDP/IP to see if that gives better performance.  My concern with TCP/IP is that error/timeout checking and retries happen at a level below where the IOC can easily keep track of what's happening.
>
> Do you plan to use VXI-11, or a raw TCP/IP stream, or a Telnet (RFC-854, RFC-2217) stream to communicate with the devices?   ASYN has support for all three options.
>
> On May 10, 2013, at 9:43 AM, Rod Nussbaumer<bomr@triumf.ca>  wrote:
>
> > Hi all.
> >
> > We are planning for installation of new beamlines requiring approximately 130 magnet power supplies. The proposed supplies (a few different manufacturers and models being considered) may be controlled using VXI-11/Telnet style interfaces, and using SCPI style commands. This implies the use of StreamDevice + Asyn interfaces in EPICS IOCs. My concern is about the scalability of ssyn with the large number of TCP connections to the power supplies. Each power supply will require an asyn IP port to be maintained. Until now, I have only configured asyn-based applications using 2 or 3 ports on a single IOC. This seems like a significant departure from what I consider a conventional system.
> >
> > The IOC platform would probably be a VME hosted CPU running Linux, with an ethernet interface dedicated as a kind of VXI-11 fieldbus. It is possible, or likely, that control of the 130 supplies will be spread across 2 or 3 hardware IOC hosts.
> >
> > Does anyone have any experience with this kind of configuration? Any problems to anticipate? Any problems that have known solutions? I am planning to try some kind of simulation using multiple physical hosts, possibly using multiple virtual hosts on each. Each host/virtual-host would simulate a power supply. Does this seem like a representative test?
> >
> > Thanks.
> >
> > Rod Nussbaumer
> > ISAC Controls, TRIUMF
> > Vancouver, Canada