Conference vaxaxp::vmsnotes

384.0. "Question on accuracy of SYS$SETTIMR" by HGOVC::EDMONDLEUNG () Wed Mar 26 1997 09:45

    Hi,
    
    I've received a mail attached below from a customer asking the time
    accuracy between the VAX and Alpha Systems. Can anyone provide
    suggestions on this issue?
    

    Subject: Question on accuracy of SYS$SETIMR.                         

    I have a question about system service SYS$SETIMR on Alpha platform:

IS THERE ANY DIFFERENCES OF SYS$SETIMR BETWEEN VAX AND ALPHA PLATFORM?

[detail] ~
Our system want to have millisecond level time accuracy for some functions, 
and we check the time period set by $SETIMR. On VAX platform, there is non 
specific problem, but when we move the code to Alpha platform, we found the 
accuracy is no longer valid.

According to the OpenVMS System Services Reference Manual, it states 
"On AXP systems, if a specified absolute time value has already passed, the 
timer expires within 10 milliseconds." How about if delta time is applied, 
what is the deviation for the timer expiration?
and "On VAX systems, if a specified absolute time value has already passed, 
the timer expires at the next clock cycle, which is within 10 milliseconds." 
I'm not sure the difference between VAX and Alpha, would you pls explain it 
more.

I have performed some tests using the system service $SETIMR on both 
platforms, the test program and results are attached, and I found the 
performance of a 3196 machine is more stable than AlphaServers 2100 and 4100, 
especially in shorter timer time period.

Hope you can give me more information and resaon about the difference of the 
system service on the two platforms. Many Thanks!
                                           
[INHERIT('sys$library:starlet.pen')]

PROGRAM test_$SETIMR (INPUT,OUTPUT);

(* AXP Jordens Temporary Debugging Use *)
[EXTERNAL,ASYNCHRONOUS] PROCEDURE LIB$PUT_OUTPUT;
EXTERN;

TYPE
  $quadword	= [UNSAFE] RECORD
			     L0	: UNSIGNED;
			     L1	: UNSIGNED;
			   END;


VAR
  time_str       : [VOLATILE] PACKED ARRAY[1..11] OF CHAR;
  sleep_time_str : PACKED ARRAY [1..40] OF CHAR VALUE '    0 00:00:00.20';
  do_time_str    : PACKED ARRAY [1..40] OF CHAR VALUE '    0 00:00:10.00';
  sleep_time     : [VOLATILE]$QUADWORD;
  do_time        : [VOLATILE]$QUADWORD;
  i_status,x     : [VOLATILE]INTEGER;

[ASYNCHRONOUS, UNBOUND]    
PROCEDURE set_timer;
BEGIN
  TIME(time_str);
  LIB$PUT_OUTPUT(%STDESCR time_str);
  i_status := $SETIMR (daytim :=sleep_time, astadr :=set_timer);
END;

BEGIN
  TIME(time_str);
  WRITELN(time_str);
  i_status := $BINTIM (timbuf:=sleep_time_str,timadr:=sleep_time);
  i_status := $BINTIM (timbuf:=do_time_str,timadr:=do_time);
  i_status := $SETIMR (daytim :=sleep_time, astadr :=set_timer);
  x := 0;
  REPEAT 
    TIME(time_str);
    WRITELN(time_str,' MAIN LOOP ',x,' sec past...');
    i_status := $SETIMR (efn:=2,daytim :=do_time);
    i_status := $WAITFR(efn:=2);
    x := x + 10;
  UNTIL x = 600; {10 minutes}
  WRITELN(i_status);
END.  

17:01:20.53
17:01:20.55 MAIN LOOP          0 sec past...
17:01:20.75
17:01:20.95
17:01:21.15
17:01:21.35
17:01:21.55
17:01:21.75
17:01:21.95
17:01:22.15
17:01:22.35
17:01:22.55
17:01:22.75
17:01:22.95
17:01:23.15
17:01:23.35
17:01:23.55
17:01:23.75
17:01:23.95
17:01:24.15
17:01:24.35
17:01:24.55
17:01:24.75
17:01:24.95
17:01:25.15
17:01:25.35
17:01:25.55
17:01:25.75
17:01:25.95
17:01:26.15
17:01:26.35
17:01:26.55
17:01:26.75
17:01:26.95
17:01:27.15
17:01:27.36
17:01:27.56
17:01:27.76
17:01:27.96
17:01:28.16
17:01:28.36
17:01:28.56
17:01:28.76
17:01:28.96
17:01:29.16
17:01:29.36
17:01:29.56
17:01:29.76
17:01:29.96
17:01:30.16
17:01:30.36
17:01:30.55 MAIN LOOP         10 sec past...
17:01:30.56
17:01:30.76
17:01:30.96
17:01:31.16
17:01:31.36
17:01:31.56
17:01:31.76
17:01:31.96
17:01:32.16
17:01:32.36
17:01:32.56
17:01:32.76
17:01:32.96
17:01:33.16
17:01:33.36
17:01:33.56
17:01:33.76
17:01:33.96
17:01:34.16
17:01:34.36
17:01:34.56
17:01:34.76
17:01:34.96
17:01:35.16
17:01:35.36
17:01:35.56
17:01:35.76
17:01:35.96
17:01:36.16
17:01:36.36
17:01:36.56
17:01:36.76
17:01:36.96
17:01:37.16
17:01:37.36
17:01:37.56
17:01:37.76
17:01:37.96
17:01:38.16
17:01:38.37
    :
    :
    :
up to 600 seconds
                                           

   I'm quite certain this topic has been raised before, and has been
   discussed at some length before.  Try a COMET search.

   Discussions of the Alpha cycle counters and the usual centisecond
   granularity of some of the major system operations is one of the
   common threads...  Check this and previous versions of this
   conference, as well as the DECC conference (I think that's where
   the cycle counters were most recently discussed), and check the
   current and previous versions of the AlphaNotes conference.)

   You will want to find out if this is a whim, or if this is a hard
   application requirement.  You will also want to find out if the
   application requires fine granularity, or high accuracy, or close
   syncrhonization -- these are not necessarily the same...