RE: dm and edm problems with CA gateway

["Kenneth Evans, Jr." <evans@aps.anl.gov>]

Rolf,

     The MEDM sliders "send on motion".  I just tried move, release, ...,
and everything else I could do to break it, but I cannot reproduce the
problem, nor, as stated, have I ever seen it.

     I should correct two things: (1) Most of our Gateways are read only to
most users, so there have probably not been extensive tests of puts through
sliders.  However, I personally use a slider (with rapid motions) on a
regular basis. (2) The Gateway is susceptible to massive puts, owing to the
round trip to the IOC.  For example, it has to queue other puts when a put
callback is in effect and wait for the response, which is much longer in
coming than in an IOC.  MEDM does not use put callbacks, but rather relies
on getting a value change event and seems to work OK.

     I still need more information, and the closer you can get to a precise
set of events that cause the problem the better.

	-Ken

-----Original Message-----
From: Rolf Keitel [mailto:rolf@triumf.ca] 
Sent: Friday, March 04, 2005 5:06 PM
To: Kenneth Evans, Jr.
Cc: tech_talk
Subject: Re: dm and edm problems with CA gateway

Hi Ken,

Kenneth Evans, Jr. wrote:

>Rolf,
>
>We have not seen such problems at the APS to my knowledge.  Our main 
>Gateway is Linux, has been running as Linux for over a year, and runs 
>at around 20,000 PVs and 2000 - 4000 events per sec.  That is, it is 
>heavily used, so one would expect to have seen this problem if it is a 
>Gateway problem.  We have on the order of 40 other Gateways, mostly 
>Solaris, running several Gateway versions.  We, of course, use MEDM.
>
>Gateway Beta18 has not been run with 3.14.6 here, only with 3.14.7.  Do 
>you have any information as to whether this happens with other Gateway 
>versions running with 3.14.6.
>  
>
we tested this with gateway 2.0.0.0Beta12 and 2.0.0.0Beta18, both on Epics
3.14.6

>You also did not say if the problem is reproducible or random.  
>Identifying a reproducible situation would be helpful.
>  
>
It does not happen every time you move the slider, but we can usually
reproduce it within 5 to 10 seconds by moving a slider a short bit, then
release, move again, release.etc.
Note that our sliders are set to "send on motion" as our operators have to
see value changes while they are moving the slider.

- rolf -

>If you can work on it yourself, you can start a Gateway on a machine or 
>port that is not used (to keep the number of events low).  You can then 
>set breakpoints in the debugger and see when these values get big.  You 
>can do the same with EDM.  At this point we do not know if EDM is 
>sending the wrong values or if the Gateway is manufacturing them.
>
>It could even be the IOC.  Is this happening with just one IOC and PV 
>or happening more universally.  Does it happen with some PVs and not
others.
>Etc.
>
>If you need help in where to set breakpoints or the like, let me know.
>There is not much else I can do from here.
>
>	-Ken
>
>  
>


-- 
+-------------------------------------------------+
| Rolf Keitel, TRIUMF         Tel: (604) 222-7453 |
| Group Leader ISAC Controls  Fax: (604) 222-7307 |
| and Electronics Development                     |
| Vancouver, B.C., Canada                         |
+-------------------------------------------------+