[ace-users] Thread stack size does not take effect
Venkatraju
I am not able to set stack size for ACE_Threads. More details in the PRF
below. Please let me know if you need any additional information.
Thanks in advance.
Venkat
ACE VERSION: 5.5.4
HOST MACHINE and OPERATING SYSTEM: i686, Linux, Fedora Core 5
TARGET MACHINE and OPERATING SYSTEM, if different from HOST:
COMPILER NAME AND VERSION (AND PATCHLEVEL): gcc version 4.1.1
20060525 (Red Hat 4.1.1-1)
THE $ACE_ROOT/ace/config.h FILE [if you use a link to a platform-
specific file, simply state which one]:
THE $ACE_ROOT/include/makeinclude/platform_macros.GNU FILE [if you
use a link to a platform-specific file, simply state which one
(unless this isn't used in this case, e.g., with Microsoft Visual
C++)]:
CONTENTS OF $ACE_ROOT/bin/MakeProjectCreator/config/default.features
(used by MPC when you generate your own makefiles):
AREA/CLASS/EXAMPLE AFFECTED:
Setting thread stack size when spawning ACE_Thread does not take effect.
DOES THE PROBLEM AFFECT:
EXECUTION of the application is affected.
SYNOPSIS:
Passing the thread stack size in ACE_Thread::spawn does not take effect.
A similar program that uses pthread_attr_setstacksize works fine.
DESCRIPTION:
Sample program below:
=======================================================
#include <iostream>
#include "ace/Thread.h"
#include "ace/Synch.h"
static int number=0;
static int seed = 0;
static void* worker(void *arg)
{
ACE_UNUSED_ARG(arg);
ACE_OS::sleep(50000);
return 0;
}
int main(int argc, char *argv[])
{
if(argc<2)
{
ACE_DEBUG((LM_DEBUG,"Usage: %s <number of threads>\n", argv[0]));
ACE_OS::exit(1);
}
const int NUM_THREADS=ACE_OS::atoi(argv[1]);
ACE_thread_t *threadID = new ACE_thread_t[NUM_THREADS];
ACE_hthread_t *threadHandles = new ACE_hthread_t[NUM_THREADS];
for (int i=0; i < NUM_THREADS; i++) {
ACE_Thread::spawn(
(ACE_THR_FUNC)worker, //entry point for new thread
0, //args to worker
THR_JOINABLE | THR_NEW_LWP, //flags
&threadID[i], // t_id
&threadHandles[i], // t_handle
ACE_DEFAULT_THREAD_PRIORITY, // priority
0, // stack
2 * 1024 * 1024 // stack size
);
}
for(int i=0; i<NUM_THREADS; i++) {
ACE_Thread::join(threadHandles[i]);
//Wait for all the threads to exit before you let the main fall through
//and have the process exit.
}
return 0;
}
=======================================================
The above test program is built and run. I monitor the stack size using:
"ps -eo pid,size,cmd | grep -i thread"
The stack size printed when run with 4 as the first argument is 41260K.
When the argument is 8, it is 82240K etc.
Wrote a similar program the uses pthread_attr_setstacksize directly.
With 4 threads, that program uses 8460K. With 8 threads, it shows
16668K as expected.
The ACE_Thread version does not seem to take stack sizes below 10MB?
Not sure if this is helpful:
"nm libACE.so.5.5.0 | grep stacksize" shows:
U pthread_attr_setstacksize
The same on my pthread test program shows:
U pthread_attr_setstacksize@@GLIBC_2.1
Pthread test program below:
========================================================
#include <pthread.h>
#include <stdio.h>
void* thr_main(void*);
int main(int argc, char *argv[])
{
int num_threads = atoi(argv[1]);
int i;
size_t stacksize;
pthread_attr_t attr;
if (pthread_attr_init(&attr) != 0) {
printf("ERROR: pthread_attr_init\n");
}
// pthread_attr_getstacksize (&attr, &stacksize);
// printf("Default stack size = %d\n", stacksize);
if (pthread_attr_setstacksize (&attr, 2*1024*1024) != 0) {
printf("ERROR: pthread_attr_setstacksize\n");
}
pthread_t thr;
for (i=0; i < num_threads; ++i) {
pthread_create(&thr, &attr, thr_main, 0);
}
pthread_join(thr, 0);
}
void* thr_main(void* arg) {
while (1) {
sleep(10);
}
}
======================================================
Douglas C. Schmidt
Hi Venkat,
Thanks for using the PRF. Can you please try replacing
ACE_Thread::spawn() with the ACE_Thread_Manager::spawn() and see if that
works better?
Thanks,
Doug
> _______________________________________________
> ace-users mailing list
> ace-...@mail.cse.wustl.edu
> http://mail.cse.wustl.edu/mailman/listinfo/ace-users
Robert Iakobashvili
On 1/9/07, Douglas C. Schmidt <sch...@dre.vanderbilt.edu> wrote:
>
> Hi Venkat,
>
> Thanks for using the PRF. Can you please try replacing
> ACE_Thread::spawn() with the ACE_Thread_Manager::spawn() and see if that
> works better?
What is your errno?
It may be a su allowed operation.
Sincerely,
Robert Iakobashvili,
coroberti %x40 gmail %x2e com
...................................................................
Navigare necesse est, vivere non est necesse
...................................................................
http://sourceforge.net/projects/curl-loader
A powerful open-source HTTP/S, FTP/S traffic
generating, loading and testing tool.
Venkatraju
On 1/9/07, Douglas C. Schmidt <sch...@dre.vanderbilt.edu> wrote:
>
> Hi Venkat,
>
> Thanks for using the PRF. Can you please try replacing
> ACE_Thread::spawn() with the ACE_Thread_Manager::spawn() and see if that
> works better?
I see the same behaviour with ACE_Thread_Manager as well. I am looking at:
http://www.google.com/codesearch?hl=en&q=+pthread_attr_setstacksize+glibc+show:LJH_-2h738s:nT0XzExAMNQ:Pr2otuY6cWA&sa=N&cd=1&ct=rc&cs_p=http://gentoo.osuosl.org/distfiles/glibc-2.3.5.tar.bz2&cs_f=glibc-2.3.5/nptl/pthread_attr_setstacksize.c#a0
Wonder if the versioned symbol part has something to do with this? And
the output
from "nm" I mentioned in my last mail:
"nm libACE.so.5.5.0 | grep stacksize" shows:
U pthread_attr_setstacksize
The same on my pthread test program shows:
U pthread_attr_setstacksize@@GLIBC_2.1
I built ACE using the standard configure, make, make install on FC5. I
also noticed
the same problem sometime back on FC4 as well. Anyone else see the same problem
on FC4/FC5?
Thanks,
Venkat
Venkatraju
On 1/9/07, Robert Iakobashvili <coro...@gmail.com> wrote:
> Hi,
>
> On 1/9/07, Douglas C. Schmidt <sch...@dre.vanderbilt.edu> wrote:
> >
> > Hi Venkat,
> >
> > Thanks for using the PRF. Can you please try replacing
> > ACE_Thread::spawn() with the ACE_Thread_Manager::spawn() and see if that
> > works better?
>
> What is your errno?
> It may be a su allowed operation.
Spawning the thread succeeds - errno is not set. I disabled selinux as well
and ran it as root and still see the same problem. Also, the pthread program
runs just fine even as a regular user.
Thanks,
Venkat
Krishnakumar B
On Tue, 09 Jan 2007 09:28:29 AM -0600, Venkatraju wrote:
> Hi,
>
> I am not able to set stack size for ACE_Threads. More details in the PRF
> below. Please let me know if you need any additional information.
[...]
> The above test program is built and run. I monitor the stack size using:
> "ps -eo pid,size,cmd | grep -i thread"
The "size" output formatter to ps prints out the memory size of the process
in kilobytes. Why do you think that the MEMORY SIZE of a process as
reported by ps is the same as the STACK SIZE of a SINGLE thread within that
process? Modify your program to print out the stack size within the newly
spawned thread using pthread_attr_getstacksize().
Since ACE_OS::thr_create calls into pthread_attr_setstacksize(), there
should be no difference in the results of pthread_attr_getstacksize()
output.
-kitty.
--
Krishnakumar B <kitty at dre dot vanderbilt dot edu>
Institute for Software Integrated Systems, Dept. of EECS, Vanderbilt University
Venkatraju
On 1/10/07, Krishnakumar B <ki...@dre.vanderbilt.edu> wrote:
> Hi,
>
> On Tue, 09 Jan 2007 09:28:29 AM -0600, Venkatraju wrote:
> > Hi,
> >
> > I am not able to set stack size for ACE_Threads. More details in the PRF
> > below. Please let me know if you need any additional information.
>
> [...]
>
> > The above test program is built and run. I monitor the stack size using:
> > "ps -eo pid,size,cmd | grep -i thread"
>
> The "size" output formatter to ps prints out the memory size of the process
> in kilobytes. Why do you think that the MEMORY SIZE of a process as
> reported by ps is the same as the STACK SIZE of a SINGLE thread within that
> process? Modify your program to print out the stack size within the newly
> spawned thread using pthread_attr_getstacksize().
>
> Since ACE_OS::thr_create calls into pthread_attr_setstacksize(), there
> should be no difference in the results of pthread_attr_getstacksize()
> output.
>
> -kitty.
Ok - I wasn't quite precise in my first mail. I don't expect the output of ps
to be the stack size of one thread. When I increase the thread count by 1 in my
program, the virtual memory usage of the program increases by about 10MB.
I think most of this is the stack size for the new thread - so I am trying to
see why that does not take effect.
Quoting from my first mail:
> The stack size printed when run with 4 as the first argument is 41260K.
> When the argument is 8, it is 82240K etc.
So, that is ~40M increase for 4 threads. I think that is because of a 10M
increase PER THREAD.
Is there ACE API for getting a running thread's stack size?
pthread_attr_getstacsize
needs a pthread_attr_t. Please let me know and I will try that.
Thanks,
Venkat
erta...@gmail.com
I have mentioned this problem way back with ACE version 5.4.10.
I am currently using ACE 5.5 and Fedora 4. (ACE installed with
configure, make, make install )
The problem still exists and is exactly the same as yours.
I hope this is resolved as soon as possible.
Ertan Deniz
Douglas C. Schmidt
The fastest way for this to get resolved is for you to step through
the source code in a debugger and figure out what's going on. If you
find the source of the problem is in ACE please send us patches to fix
it.
Thanks,
Doug
--
Dr. Douglas C. Schmidt Professor and Associate Chair
Electrical Engineering and Computer Science TEL: (615) 343-8197
Vanderbilt University WEB: www.dre.vanderbilt.edu/~schmidt
Nashville, TN 37203 NET: d.sc...@vanderbilt.edu
Steve Huston
> From: ace-user...@cse.wustl.edu
> [mailto:ace-user...@cse.wustl.edu] On Behalf Of Douglas
> C. Schmidt
> Sent: Thursday, February 15, 2007 8:24 AM
> To: erta...@gmail.com; ace-...@cse.wustl.edu
> Subject: Re: [ace-users] Thread stack size does not take effect
>
>
> Hi Ertan,
>
> >I have mentioned this problem way back with ACE version 5.4.10.
> >
> >http://groups.google.com.tr/group/comp.soft-sys.ace/browse_th
read/thread/4a34783c99a25ed2/4f08b47d7bc9fa3b?lnk=gst&q=user+thread+st
ack&rnum=6&hl=tr#4f08b47d7bc9fa3b
>
> >I am currently using ACE 5.5 and Fedora 4. (ACE installed with
> >configure, make, make install )
> >The problem still exists and is exactly the same as yours.
>
> >I hope this is resolved as soon as possible.
>
> The fastest way for this to get resolved is for you to step through
> the source code in a debugger and figure out what's going on. If
you
> find the source of the problem is in ACE please send us patches to
fix
> it.
... Or take advantage of the services of one of the ACE support
companies who'd be more than happy to relieve you of that debugging
effort.
-Steve
--
Steve Huston, Riverace Corporation
New Support options to help you with ACE!
See http://www.riverace.com/support.htm
erta...@gmail.com
> > > The ACE_Thread version does not seem to takestacksizes below 10MB?
>
> > > Not sure if this is helpful:
> > > "nm libACE.so.5.5.0 | grep stacksize" shows:
> > > U pthread_attr_setstacksize
>
> > > The same on my pthread test program shows:
> > > U pthread_attr_setstacksize@@GLIBC_2.1
>
By consentrating on this info I finally managed to identify and solve
the problem.
ACE library was strangely calling an older pthread_create which
assumed
that pthread_attr_t does not contain stacksize etc...
It was simply ignoring the stacksize attribute.
The problem is in the link stage. In the link command line -lpthread
must come after the object files (or .a libraries) that
use them on the link line.
The solution came from the following places;
https://mail.gna.org/public/xenomai-core/2005-12/msg00034.html
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=145941
The only thing we need to do is change the file:
ACE_wrappers/ace/Makefile.am
Old LDFLAGS:
libACE_la_LDFLAGS = \
-release @ACE_VERSION_NAME@
New LDFLAGS:
libACE_la_LDFLAGS = \
-release @ACE_VERSION_NAME@ -lpthread
Tried and verified that passing the stacksize argument works.
Sorry that I couldn't send this as a diff patch.
Ertan Deniz