Experimental Physics and
Industrial Control System

"Konrad, Martin via Tech-talk" <tech-talk at aps.anl.gov> · Tue, 15 Sep 2020 18:27:08 +0000

Hi Jalal,
> I want to ask you if like to see a performance evaluation
> paper/report of the CA SoftIOC.
>
> If so, what parameters you would like to see in such a report? E.g.
> I can suggest the maximum number of records per instance, the
> highest attainable frequency...
Ralph is right, predicting performance can be quite difficult since
operating system, CPU features, cache size etc. might have a significant
impact on IOC performance. Plus in a real-world scenario you might need
to talk to hardware - which introduces latency, locks etc. making things
even harder to predict. Simply measuring resource consumption for one
device and extrapolating to the number of devices you plan to use might
work better. Keep in mind that lock contention might become more
relevant the more threads/cores you add, though. Generally
virtualization is your friend here: If your estimates occasionally
aren't accurate you can easily adjust resources.

These problems aside: I'm running a couple of IOCs hosting several
hundred thousand records each. In practice we are burning the majority
of our CPU/RAM resources for Channel Access connections. Our archiver
establishing connections to ~30% of our records is by far the biggest
driver here (resources for handling CA connections often exceed the
resources the IOC needs to communicate with its devices). With this in
mind I would suggest to take Channel Access connections into account
when evaluating IOC performance.

Cheers,

Martin

-- 
Martin Konrad
Facility for Rare Isotope Beams
Michigan State University
640 South Shaw Lane
East Lansing, MI 48824-1321, USA
Tel. 517-908-7253
Email: konrad at frib.msu.edu

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System