Experimental Physics and Industrial Control System
Folks,
I just wanted to follow up on Jeff Hill's post regarding my observations on
the time required for ca_pend_event().
There is an error in the R3.12 Channel Access Reference Manual. In the
comments under the description of ca_pend_event() it says that:
"A very short TIMEOUT such as .0001 seconds will result in a poll
(i.e. channel access will just check for file descriptor activity and
then return)."
This is incorrect. In fact the threshold is 1 microsecond. Only TIMEOUT
values below 1 microsecond result in a poll. TIMEOUTs of 2 microseconds will
often delay ca_pend_event by 1 tick of the system clock. Measured values of
the time required for ca_pend_event(2.e-6) are 10 milliseconds on Sun4 and
SGI, 1 msec on Alpha, <.4 msec on Solaris.
Jeff has suggested 2 solutions to the problem (in addition to fixing the
documentation):
- Setting the threshold to (1.0/CLOCKS_PER_SEC). Since CLOCKS_PER_SEC is ansi
this would be portable. Note that, for example, this would result in any
TIMEOUT less than 10 msec on a Sun4 result in a poll.
- Making the TIMEOUT value which will result in a poll a symbolic constant.
Note that the value of SELECT_POLL, defined in iocinf.h cannot be used,
since it is defined as 25 msec.
I think that both of the above suggestions are good ones and should be
implemented.
The performance impact of using the correct values for TIMEOUT can be very
large. A factor of 5 improvement in the performance of 1000 calls to
ezcaGet() was observed when SHORT_TIME in ezca was redefined from 10
microseconds to 1 nanosecond.
____________________________________________________________
Mark Rivers (312) 702-2279 (office)
CARS (312) 702-9951 (secretary)
Univ. of Chicago (312) 702-5454 (FAX)
5640 S. Ellis Ave. (708) 922-0499 (home)
Chicago, IL 60637 [email protected] (Internet)
- Replies:
- Re: ca_pend_event() Jeff Hill
- Navigate by Date:
- Prev:
Re: Bug in OMS Driver routines Guy Jennings
- Next:
Device support for MKS vacuum gauges? Mark Rivers
- Index:
1994
1995
<1996>
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
- Navigate by Thread:
- Prev:
more vxWorks statistics Jim B. Kowalkowski
- Next:
Re: ca_pend_event() Jeff Hill
- Index:
1994
1995
<1996>
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024