Experimental Physics and
Industrial Control System

Frank Lenkszus <[email protected]> · Thu, 22 Jul 1999 09:12:22 -0500 (CDT)

I've made some channel access response time measurements between our OPI 
workstations and our storage-ring bpm iocs. The measurement does cagets to
PVs in the bpm iocs. The time from issuance of the caget to the received PV
value is measured. This is not the same as the it takes for a CA client to
receive a monitor, but it's a lot easier to measure. I bin the results
in a histogram with a 100msec resolution.

The response time depends on ioc load, workstation load and network load.
Most response times are less than 100ms. About 0.25 to 1% of the time the
response time is 100 to 200 ms. Occasionally the response time exceeds 200ms.
I've seen response times out to 500-600 ms, but this is rare. Keep in mind
this measurement is UNIX<->vxWorks. I would expect (hope) vxWorks<->vxWorks
to be faster. The Unix workstations are Sun Ultras running Solaris. The iocs
are running vxWorks 5.3 on MVME172s (60MHz). The iocs are running about 50 to
70% idle. The workstations are typically busy. Sorry, I didn't note the 
workstation load at the time of the measurement.

Frank

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System