Experimental Physics and
Industrial Control System

"jun-ichi.odagiri--- via Tech-talk" <tech-talk at aps.anl.gov> · Tue, 30 Mar 2021 12:49:16 +0900

Dear Ralph,

I make a correction.

> What I wanted to do was to give "CAS-client" a priority
> higher than that of "scan-0.1" because commands sent from
> a CA client to an IOC start to be lost when the number of
> records being processed with the SCAN value of ".1 seconds"
> was increased.

The problem was not loss of CA messages but some funny
behavior of the device under EPICS control. I made many
mistakes since I'm now very impatient and confused because
the deadline is approaching very close.

To all.
I apologize to all of you if "you" in Ralph's mail does
not mean me.

J.Odagiri at KEK in a hurry

----- Original Message -----
> Dear Ralph,
> 
> I assume that "you" means me in your message because I
> triggered this chain of e-mails. If not, please just ignore
> this e-mail.
> 
> (I know "you" can mean unspecified persons in English.)
> 
> What I wanted to do was to give "CAS-client" a priority
> higher than that of "scan-0.1" because commands sent from
> a CA client to an IOC start to be lost when the number of
> records being processed with the SCAN value of ".1 seconds"
> was increased.
> 
> But now I understand that the problem of mine is not an
> issue of priorities of threads but an issue of something
> else.
> 
> (I think default priorities of the EPICS IOC threads
> are well-chosen as you mentioned in a previous e-mail.)
> 
> I do not want to choose multiple IOCs solution because
> I want to keep the system as simple as possible.
> 
> The value of PRIO field is irrelevant because I do not
> rely on callback threads in my device support.
> 
> I can not choose multiple parallel callback solution
> because my system is single-CPU system.
> 
> I now suspect that the origin of my problem is not on
> IOC-side but on a custom device-side being controlled
> by the IOC.
> 
> Lots of thanks anyway for your detailed comments and
> advice.
> 
> Best regards,
> 
> J.Odagiri at KEK
> 
> 
> ----- Original Message -----
> > So, coming back to what (I think) was the original question:
> > How to change the priority of the general purpose callback threads (
> named
> > "cb...") to be higher than a periodic scanning thread?
> > 
> > You don't have to. Use the PRIO field.
> > There are three (sets of) general purpose callback threads, related 
to 
> the
> > three values of the records' PRIO field: Low, Medium and High.
> > Their priorities are designed to be lower than all periodic 
processing
> > (Low), halfway between the slow and the fast periodic processing (
> Medium)
> > and higher than all period processing (High). For each record, the 
> thread
> > used for callback processing is selected based on the setting of its 
> PRIO
> > field.
> > On multi-CPU systems, you can also enable multiple parallel callback
> > threads (e.g., one per CPU) to boost the callback and 'I/O Intr' 
> processing
> > power of your IOC.
> > 
> > Only if these mechanisms show not being sufficient for your 
> application,
> > additional steps need to be taken. In most cases, I would assume 
> further
> > parallelization (splitting up the application to multiple IOCs, 
adding 
> more
> > CPU power) to be more efficient than tuning priorities.
> > 
> > Cheers,
> > ~Ralph
> > 
> 
> 
> 

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System