Re: RTLinux for an IOC

[Marty Kraimer <mrk@aps.anl.gov>]

Richard.Farnsworth@mpv.vic.gov.au wrote:
> G'day,
>
> There is a great work going on with RTEMS and, of course Tornado (vxWorks)
> is the traditional IOC, but is anyone using a semi-commercial variant of RT
> Linux as a serious IOC? For example I'm looking at RedHawk Linux
> (Concurrent computers -
> http://www.ccur.com/isd_solutions_redhawklinux.asp).
>
> Any fundamental reason why this might not be a good idea before I try it?
> Any experiences or thoughts ?


As I recall RTLinux is a small real-time kernel that also runs the regular Linux 
kernel as a  low priority RTLinux thread. Since an integral part of EPICS is 
Channel Access it is hard to see how an IOC could be run using only RTLinux threads.

Once the 2.6 kernel is available, linux will have much better real time 
characteristics. My guess is that, with more work needed on 
base/src/libCom/osi/(posix and/or linux), linux will be sufficient for most 
EPICS IOC applications.

For applications with harder real time requirements at least two other 
approaches are available. Both require that the real time requirements are 
localized to a small part of the application. If the entire application has hard 
real-time requirements than perhaps EPICS is the wrong solution.

On method is to use RTLinux or RTAI for the hard real-time and write EPICS 
device/driver support that communicates with the RTLinux or RTAI threads.

Another approach is becoming available. I/O boards are becoming available that 
include easily programmable FPGAs and high performance Digital and Analog I/O. 
The FPGAs can be programmed to handle the hard real-time and again EPICS 
device/driver support can be created to communicate with the FPGA code.

Just my thoughts.

Marty Kraimer