[ace-users] Innacurate timers in Linux after 5.5.6

[Johnny Willemsen]

Hi,

Have you also tested with the traditional way of compilation? My first
question would be whether this is a code change or an autoconf change you
see. Be aware of the fact that 99% of the tests on the scoreboard is using
traditional way of compilation.

Regards,

Johnny Willemsen
Remedy IT
Postbus 101
2650 AC  Berkel en Rodenrijs
The Netherlands
www.theaceorb.nl / www.remedy.nl  

*** Integrated compile and test statistics see
http://scoreboard.theaceorb.nl ***
*** Commercial service and support for ACE/TAO/CIAO             ***
*** See http://www.theaceorb.nl/en/support.html                 ***

> -----Original Message-----
> From: ace-users-bounces at cse.wustl.edu 
> [mailto:ace-users-bounces at cse.wustl.edu] On Behalf Of Alberto 
> Salmerón Moreno
> Sent: Monday, July 23, 2007 8:10 PM
> To: ace-users at cs.wustl.edu
> Subject: [ace-users] Innacurate timers in Linux after 5.5.6
> 
> Hi all,
> 
> I've observed that, using a patched Linux kernel to better support
> real-time requirements, newer versions of ACE fail to achieve the
> expected, fine-grained timer resolution while older ones work as
> expected. PRF follows.
> 
> ACE VERSION: 5.5.9
> 
> HOST MACHINE and OPERATING SYSTEM: Intel Pentium III, RedHat Linux 9,
> 2.4.20-rxv kernel (patches from
> http://www.rexursive.com/software/redhat.html)
> 
> TARGET MACHINE and OPERATING SYSTEM, if different from HOST:
> 
> COMPILER NAME AND VERSION (AND PATCHLEVEL): 3.2.2 20030222 (Red Hat
> Linux 3.2.2-5)
> 
> THE $ACE_ROOT/ace/config.h FILE: I used the configure script,
> resulting config.h is 64Kb. I will post it if needed
> 
> THE $ACE_ROOT/include/makeinclude/platform_macros.GNU FILE:
> 
> CONTENTS OF $ACE_ROOT/bin/MakeProjectCreator/config/default.features:
> 
> AREA/CLASS/EXAMPLE AFFECTED:
> 
> DOES THE PROBLEM AFFECT: EXECUTION of application
> 
> SYNOPSIS:
> The average dispatch time for a repeated timer scheduled on a 
> reactor is
> different from the expected time, while on a previous version of ACE
> (5.5.6) it was right.
> 
> DESCRIPTION:
> During the development of a application with severe time requirements
> I've noticed that timer resolution on Linux seems to have decreased
> since ACE version 5.5.6. The application needs a timer to be
> dispatched every 2ms. With version 5.5.6 this requirement was
> attained, but on the 5.5.8 and 5.5.9 releases the timer was activated
> every 12ms. I plan to test with versions 5.5.7 and 5.5 as well, to
> further locate the source of this problem.
> 
> Since this test is run on the same machine and OS, but with different
> versions of ACE, I guess that this is a problem within ACE.
> 
> I would greatly appreciate any insight on this, or an indication of
> where in the ACE source code can I find the timer handling code so I
> can try to figure this out myself.
> 
> Note: This can be considered as a follow up to my previous post
> (http://groups.google.com/group/comp.soft-sys.ace/browse_threa
d/thread/4da3a151b47d70a4/b8e088e24bd2a3ef?lnk=gst&rnum=1),
> but using a more suitable kernel and then changing the version of ACE,
> confirming that on older versions of ACE timing works better.
> 
> REPEAT BY:
> 
> Sample program follows. A timer fires every 2ms, 60 times. Then the
> total time taken is printed on the screen, which should be roughly
> 2*60ms = 120ms.
> 
> /* test.cpp - start *********************/
> #include "ace/High_Res_Timer.h"
> #include "ace/Log_Msg.h"
> #include "ace/OS.h"
> #include "ace/OS_NS_Thread.h"
> #include "ace/Reactor.h"
> #include "ace/Timer_Heap.h"
> #include "ace/Timer_Queue.h"
> 
> class TimerHandler : public ACE_Event_Handler {
>     public:
>         int n;
>         long i;
>         ACE_Time_Value t_end;
> 
>     	TimerHandler(int n_) : n(n_), i(0), 
> t_end(ACE_Time_Value::zero) {};
>     		
>         int handle_timeout(const ACE_Time_Value &ti, const 
> void * = 0) {
>             if (i == n) {
>                 t_end = ti;
>                 reactor()->end_reactor_event_loop();
>                 return -1;
>             }
>             else {
>                 i++;
>                 return 0;
>             }
>         }
> };
> 
> int ACE_TMAIN(int argc, ACE_TCHAR *argv[]) {
>     int time = argc > 1 ? ACE_OS::atoi(argv[1]) : 2000; // 2ms
> 
>     ACE_Sched_Priority priority = 1;
>     if (ACE_OS::thr_setprio (priority) == -1) {
>     	ACE_ERROR ((LM_ERROR, "(%P|%t) main thr_setprio failed,"
>         		"no real-time features\n"));
>     }
> 
>     ACE_High_Res_Timer::global_scale_factor();
>     ACE_Reactor reactor;
>     
> reactor.timer_queue()->gettimeofday(&ACE_High_Res_Timer::getti
meofday_hr);
>     ACE_Time_Value initialDelay = ACE_Time_Value(0, 0);
>     ACE_Time_Value interval (0, time);
>     TimerHandler *handler = new TimerHandler(60);
>     if (reactor.schedule_timer(handler, 0, initialDelay, 
> interval) == -1) {
>     	ACE_ERROR_RETURN ((LM_ERROR, "%p\n", "schedule_timer()"), 1);
>     }
>     ACE_Time_Value t_start = reactor.timer_queue()->gettimeofday();
>     if (reactor.run_reactor_event_loop() == -1) {
>     	ACE_ERROR_RETURN ((LM_ERROR, "%p\n", 
> "run_reactor_event_loop()"), 1);
>     }
>     ACE_Time_Value t_diff = handler->t_end - t_start;
>     ACE_DEBUG((LM_DEBUG, "%ds %dus\n", t_diff.sec(), t_diff.usec()));
>     return 0;
> }
> /* test.cpp - end ***********************/
> 
> SAMPLE FIX/WORKAROUND: Use an older version of ACE (e.g. 5.5.6).
> 
> --
> Alberto Salmerón Moreno
> Departamento de Lenguajes y Ciencias de la Computación
> Universidad de Málaga, ETSI Informática
> salmeron at lcc.uma.es
> 
> _______________________________________________
> ace-users mailing list
> ace-users at mail.cse.wustl.edu
> http://mail.cse.wustl.edu/mailman/listinfo/ace-users
>