Experimental Physics and
Industrial Control System

Andrew Johnson <[email protected]> · Thu, 20 Oct 2005 17:01:39 -0500

Kay,

I want to make sure you're considering one point regarding deadbands:

Kay-Uwe Kasemir wrote:

* update on change in value
 with relative and absolute deadbands,
 each as a positive and negative limit.
 E.g. update when value increases by >= 1% or descreases by <=0.5%.
 EPICS V3 only has a single absolute deadband.

...

* update on archive related changes
 which are designated relative and absolute deadbands,
 again pos. & neg.,
 and in addition a minimum update rate.
 EPICS V3 only ADEL, no additional periodic update.

"Designated" by who? In V3 the ADEL and MDEL fields allows the DB

designer to set appropriate per-record deadbands; just allowing every

client to set its own deadband in the subscription request doesn't

provide any way for the DB designer to even hint what an appropriate

deadband might be, which could signficantly impact IOC performance.

Maybe that information doesn't belong in the database at all, but I'm

not convinced of that; I don't think we have all the requirements for V4

deadbands specified yet.

Here's one possible scenario: Say I have an IOC with 10,000 analog input

points being read from sources which normally only change very slowly,

but are subject to noise. Each point is read at 1 Hz because the system

needs to know about any fast changes that might occur, but with the

correct deadbands I'm only expecting 100 changes per second. Without

any deadbands though the IOC would actually send out 10,000 changes per

second, which would not be good. Each channel is only being updated at

a slow 1 Hz rate, and I imagine most display managers will probably

subscribe with a rate limit higher than that, so a single naive display

screen could overload my IOC.

As a database designer, I want a way to protect the IOC from naive

screen designers who forget about things like deadbands. In V3 I could

set the MDEL and ADEL fields on each record according to the noise on

that channel, and I wouldn't have to worry about clients getting deluged

with data. The deadband was also configured centrally in the IOC, so

none of the clients had to know anything about the channels they connect

to other than their name.

I guess I'm suggesting that the database should still play some role in

setting deadbands, maybe by providing values that general purpose

clients like MEDM and the archiver will use by default, although they

should be able to override the IOC value.

- Andrew
--
English probably arose from Normans trying to pick up Saxon girls.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System