Odd conference memory corruption issue
I'm not entirely sure PJSIP is causing this, but here is what's
happening:
- I establish a SIP call somewhere.
- Create a recorder instance, connect to the call and sound input
device. - Mute the audio (i.e. the main input device is disconnected from
the recorder & call.) - At this point, memory corruption happens constantly in the
background. - Unmuting (i.e. reconnecting the sound input device to the call and
recorder) stops the memory corruption from continuing.
Any ideas on how to track this one down?
Norman Franke
ASD, Inc.
More on this issue. If I just use the mute switch on my headset, I
can reproduce the crash. It appears that PJSIP is doing something odd
if it detects silence from the mic.
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:04 PM, Norman Franke wrote:
I'm not entirely sure PJSIP is causing this, but here is what's
happening:
- I establish a SIP call somewhere.
- Create a recorder instance, connect to the call and sound input
device. - Mute the audio (i.e. the main input device is disconnected from
the recorder & call.) - At this point, memory corruption happens constantly in the
background. - Unmuting (i.e. reconnecting the sound input device to the call
and recorder) stops the memory corruption from continuing.
Any ideas on how to track this one down?
Norman Franke
ASD, Inc.
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
I have no_vad set to true, and it still happens if I set ec_tail_len
= 0. Setting no_vad = false still causes the problem to appear. Is
there some other silence detection in pjmedia?
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:38 PM, Norman Franke wrote:
More on this issue. If I just use the mute switch on my headset, I
can reproduce the crash. It appears that PJSIP is doing something
odd if it detects silence from the mic.
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:04 PM, Norman Franke wrote:
I'm not entirely sure PJSIP is causing this, but here is what's
happening:
- I establish a SIP call somewhere.
- Create a recorder instance, connect to the call and sound input
device. - Mute the audio (i.e. the main input device is disconnected from
the recorder & call.) - At this point, memory corruption happens constantly in the
background. - Unmuting (i.e. reconnecting the sound input device to the call
and recorder) stops the memory corruption from continuing.
Any ideas on how to track this one down?
Norman Franke
ASD, Inc.
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
I tried to reproduce it here (on a Windows machine) but it didn't
happen, unfortunately. So maybe this behavior is specific to PA
implementation on Mac?
Which part of memory got corrupted? And where the crash is located?
cheers,
-benny
Norman Franke wrote:
I have no_vad set to true, and it still happens if I set ec_tail_len =
0. Setting no_vad = false still causes the problem to appear. Is there
some other silence detection in pjmedia?
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:38 PM, Norman Franke wrote:
More on this issue. If I just use the mute switch on my headset, I can
reproduce the crash. It appears that PJSIP is doing something odd if
it detects silence from the mic.
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:04 PM, Norman Franke wrote:
I'm not entirely sure PJSIP is causing this, but here is what's
happening:
- I establish a SIP call somewhere.
- Create a recorder instance, connect to the call and sound input
device. - Mute the audio (i.e. the main input device is disconnected from
the recorder & call.) - At this point, memory corruption happens constantly in the
background. - Unmuting (i.e. reconnecting the sound input device to the call and
recorder) stops the memory corruption from continuing.
Any ideas on how to track this one down?
Norman Franke
ASD, Inc.
It's somewhat random, but the memory that gets corrupted is in
unallocated memory. The OS X malloc checks checksums on freed memory
and if it was modified after free, it will complain next malloc. This
is quite reproducible for me. I can't imagine it's PA, since setting
the conference level to zero causes the same problem, so the PA
implementation is sill getting good audio. Whether I disconnect the
sound input from the bridge or set the volume to zero, the bug appears.
Speaking of which, does PJSIP do some sort of gain control? We are
having issues what something doing gain control that amplifies
background noise when a client isn't talking. I'm not sure if
Asterisk is doing it, PJSIP or PA.
Norman Franke
ASD, Inc.
On Jan 4, 2008, at 2:05 AM, Benny Prijono wrote:
I tried to reproduce it here (on a Windows machine) but it didn't
happen, unfortunately. So maybe this behavior is specific to PA
implementation on Mac?
Which part of memory got corrupted? And where the crash is located?
cheers,
-benny
Norman Franke wrote:
I have no_vad set to true, and it still happens if I set
ec_tail_len =
0. Setting no_vad = false still causes the problem to appear. Is
there
some other silence detection in pjmedia?
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:38 PM, Norman Franke wrote:
More on this issue. If I just use the mute switch on my headset,
I can
reproduce the crash. It appears that PJSIP is doing something odd if
it detects silence from the mic.
Norman Franke
ASD, Inc.
On Jan 3, 2008, at 6:04 PM, Norman Franke wrote:
I'm not entirely sure PJSIP is causing this, but here is what's
happening:
- I establish a SIP call somewhere.
- Create a recorder instance, connect to the call and sound input
device. - Mute the audio (i.e. the main input device is disconnected from
the recorder & call.) - At this point, memory corruption happens constantly in the
background. - Unmuting (i.e. reconnecting the sound input device to the
call and
recorder) stops the memory corruption from continuing.
Any ideas on how to track this one down?
Norman Franke
ASD, Inc.
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Norman Franke wrote:
It's somewhat random, but the memory that gets corrupted is in
unallocated memory. The OS X malloc checks checksums on freed memory and
if it was modified after free, it will complain next malloc. This is
quite reproducible for me. I can't imagine it's PA, since setting the
conference level to zero causes the same problem, so the PA
implementation is sill getting good audio. Whether I disconnect the
sound input from the bridge or set the volume to zero, the bug appears.
The interesting thing is this seems to only happen if the signal
from the microphone is muted, either by muting the microphone
(physically) or disconnecting the mic in the bridge (is it?). I'm
having different behavior here with the same scenario. Here, instead
of memory corruption, my playback (of the call) gets garbled if I
disconnect the mic input, but this only happens with Speex and not
with other codecs. I'm still working on this issue, and who knows
maybe it's related to yours.
What codec are you using?
Speaking of which, does PJSIP do some sort of gain control? We are
having issues what something doing gain control that amplifies
background noise when a client isn't talking. I'm not sure if Asterisk
is doing it, PJSIP or PA.
The only gain control in PJSIP is in the bridge. If you never call
pjsua/pjmedia_conf_adjust_tx/rx_level() then no gain control is
done. Having said that, the mixing functionality in the bridge also
inherently adjusts the gain of the inputs, if more than one inputs
are mixed together. When one sink output only receives from one
input, the input signal will pass through unmodified.
cheers,
-benny
Norman Franke
ASD, Inc.
Hi,
I'm using 0.5.10.4. When I send STUN request, a error
(PJLIB_UTIL_ESTUNNOTRESPOND) usually returned. But I'm sure the STUN server
has respond to client within 200ms. I use Ethereal to capture network
packets. I don't understand why PJSIP usually return
(PJLIB_UTIL_ESTUNNOTRESPOND)?
It's somewhat random, but the memory that gets corrupted is in
unallocated memory. The OS X malloc checks checksums on freed
memory and
if it was modified after free, it will complain next malloc. This is
quite reproducible for me. I can't imagine it's PA, since setting the
conference level to zero causes the same problem, so the PA
implementation is sill getting good audio. Whether I disconnect the
sound input from the bridge or set the volume to zero, the bug
appears.
The interesting thing is this seems to only happen if the signal
from the microphone is muted, either by muting the microphone
(physically) or disconnecting the mic in the bridge (is it?). I'm
having different behavior here with the same scenario. Here, instead
of memory corruption, my playback (of the call) gets garbled if I
disconnect the mic input, but this only happens with Speex and not
with other codecs. I'm still working on this issue, and who knows
maybe it's related to yours.
What codec are you using?
I'm using PCMU. (Using it over a 100 Mbps LAN, I don't see a point in
compressing.)
Speaking of which, does PJSIP do some sort of gain control? We are
having issues what something doing gain control that amplifies
background noise when a client isn't talking. I'm not sure if
Asterisk
is doing it, PJSIP or PA.
The only gain control in PJSIP is in the bridge. If you never call
pjsua/pjmedia_conf_adjust_tx/rx_level() then no gain control is
done. Having said that, the mixing functionality in the bridge also
inherently adjusts the gain of the inputs, if more than one inputs
are mixed together. When one sink output only receives from one
input, the input signal will pass through unmodified.
I was thinking more dynamic or automatic gain control? We are
struggling with background noise, and it seems something is
amplifying it when the client's user isn't talking. In this case, it
should just be the remote side going to the headset.
-Norman
On 1/7/08, Mimic hm super_mimic@hotmail.com wrote:
Hi,
I'm using 0.5.10.4. When I send STUN request, a error
(PJLIB_UTIL_ESTUNNOTRESPOND) usually returned. But I'm sure the STUN
server
has respond to client within 200ms. I use Ethereal to capture network
packets. I don't understand why PJSIP usually return
(PJLIB_UTIL_ESTUNNOTRESPOND)?
Please use at least version 0.8 or better the SVN version. Your copy is more
than 9 months old and is no longer supported.
cheers,
-benny
On 1/7/08, Norman Franke norman@myasd.com wrote:
What codec are you using?
I'm using PCMU. (Using it over a 100 Mbps LAN, I don't see a point in
compressing.)
I've just committed this ticket (http://www.pjsip.org/trac/ticket/439), can
you try it out?
cheers,
-benny
It seems to be fine now, although it was a bit random (and my app has
changed since then.) Did this bug affect PCMU as well?
Norman Franke
ASD, Inc.
On Jan 7, 2008, at 8:15 PM, Benny Prijono wrote:
On 1/7/08, Norman Franke norman@myasd.com wrote:
What codec are you using?
I'm using PCMU. (Using it over a 100 Mbps LAN, I don't see a point in
compressing.)
I've just committed this ticket ( http://www.pjsip.org/trac/ticket/
439), can you try it out?
cheers,
-benny
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at 1.0
and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than not.
Any thoughts on this one?
Norman Franke
ASD, Inc.
On Jan 8, 2008, at 1:28 PM, Norman Franke wrote:
It seems to be fine now, although it was a bit random (and my app
has changed since then.) Did this bug affect PCMU as well?
Norman Franke
ASD, Inc.
On Jan 7, 2008, at 8:15 PM, Benny Prijono wrote:
On 1/7/08, Norman Franke norman@myasd.com wrote:
What codec are you using?
I'm using PCMU. (Using it over a 100 Mbps LAN, I don't see a point in
compressing.)
I've just committed this ticket ( http://www.pjsip.org/trac/ticket/
439), can you try it out?
cheers,
-benny
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Hi everybody,
I'm experiencing some trouble here with high cpu usage when I compile
my application for a large number of calls, and I would like to try
using epoll. I don't know if this would help, but it may be a good
idea to try it out. I tried setting export LINUX_POLL := epoll inside
build.mak, but It didn't use epoll. I've already set the following
parameters:
media_cfg.ec_tail_len = 0;
media_cfg.quality = 1;
media_cfg.clock_rate = 8000;
media_cfg.no_vad = PJ_TRUE;
debug level is zero, and I compiled both the api and my application
without a debug flag.
using the "top" application (I'm using linux) I could see the thread
that is using the CPU power, it has a PID one number bigger than the
first thread (main application thread). I didn't find a way to get
the name of the thread inside the application just with the pid, but
I'm pretty sure it should be the first thread created by application,
and I don't know which one is it(pjsua worker thread, media,
clock...). with this I may reduce the scope to find the problem. I
don't think It has to do with my code, since I made a new application
that just receives a call and plays a wav when media is ready and it
gets a high cpu usage also. If I start my application with a small
number in pjsua_config::max_calls it has no problems, but if I put a
large number like 500 calls for example, the cpu usage goes high with
as little as 5 calls. I tested the same code against version 0.7.0 of
the api and the same problem happens. My current code is running in
api r1629. The cpu usage is not constant, it comes in bursts of a few
seconds and then some time at a low usage (shows 0% in "top"). if I
increase the number of call the amount of time of low cpu usage
reduces until it disappears with about 20 concurrent calls.
I did some tests with sipp and the cpu usage is low, I don't really
know why, maybe it doesn't send any audio packets or something else.
All my test call that showed the problem where made from softphones
in other computers, different versions too (twinkle in linux and
xlite in windows). I tested the application in different machines,
all running linux.
I'm guessing that the use of epoll may help because I talked about
this problem a long time ago in the list and some people using
windows reported back that they don't have the problem, so this
should be related to something specific to linux, but I'm not really
sure here.
thanks for the great job in the api,
Thiago
Abra sua conta no Yahoo! Mail, o único sem limite de espaço para armazenamento!
On 1/8/08, Norman Franke norman@myasd.com wrote:
It seems to be fine now, although it was a bit random (and my app has
changed since then.) Did this bug affect PCMU as well?
I didn't think so, since the problem here seems to be specific to speex. But
now that you said your application is working, probably it did affect pcmu
as well.
-benny
Norman Franke
ASD, Inc.
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
More on this one. In conference.c line 1797, if I comment out this as
follows:
// conf_port->last_level = level;
Then don't get a click when the playback finishes. Seems like maybe
there is an issue when the number of transmitting ports drops?
-Norman
On Jan 9, 2008, at 11:35 AM, Norman Franke wrote:
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
As far as I can see, it may be when there are, say 3 sources, in the
conference being mixed, then we drop down to two. So if the "level"
is, say 1, we were dividing by 3 and now it's by 2.
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 12:16 PM, Norman Franke wrote:
More on this one. In conference.c line 1797, if I comment out this
as follows:
// conf_port->last_level = level;
Then don't get a click when the playback finishes. Seems like
maybe there is an issue when the number of transmitting ports drops?
-Norman
On Jan 9, 2008, at 11:35 AM, Norman Franke wrote:
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than
not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
More weirdness. I have the call's relative volume at 150% and the mic
normal (default.) If I record a call, the caller is way, way quieter
than the person using the microphone. Why is this?
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 2:26 PM, Norman Franke wrote:
As far as I can see, it may be when there are, say 3 sources, in
the conference being mixed, then we drop down to two. So if the
"level" is, say 1, we were dividing by 3 and now it's by 2.
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 12:16 PM, Norman Franke wrote:
More on this one. In conference.c line 1797, if I comment out this
as follows:
// conf_port->last_level = level;
Then don't get a click when the playback finishes. Seems like
maybe there is an issue when the number of transmitting ports drops?
-Norman
On Jan 9, 2008, at 11:35 AM, Norman Franke wrote:
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than
not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Bear with me on this, Norman, we're currently looking at this issue.
-benny
On 1/10/08, Norman Franke norman@myasd.com wrote:
More weirdness. I have the call's relative volume at 150% and the mic
normal (default.) If I record a call, the caller is way, way quieter than
the person using the microphone. Why is this?
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 2:26 PM, Norman Franke wrote:
As far as I can see, it may be when there are, say 3 sources, in the
conference being mixed, then we drop down to two. So if the "level" is, say
1, we were dividing by 3 and now it's by 2.
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 12:16 PM, Norman Franke wrote:
More on this one. In conference.c line 1797, if I comment out this as
follows:
// conf_port->last_level = level;
Then don't get a click when the playback finishes. Seems like maybe there
is an issue when the number of transmitting ports drops?
-Norman
On Jan 9, 2008, at 11:35 AM, Norman Franke wrote:
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes. It's
random, so it doesn't always do it, but seems to more often than not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Great! I've temporarily worked around this by ensuring only a single
output is active at the start of the call (when we play a recorded
message.) Other playbacks are still a problem, but happen much less
often.
I've then set the gain of the caller to 250% which makes the
recordings sound good.
However, I am anxious for a better fix to the conference mixing code.
If you need more info, let me know.
Norman Franke
Answering Service for Directors, Inc.
www.myasd.com
On Jan 11, 2008, at 8:01 AM, Benny Prijono wrote:
Bear with me on this, Norman, we're currently looking at this issue.
-benny
On 1/10/08, Norman Franke < norman@myasd.com> wrote:
More weirdness. I have the call's relative volume at 150% and the
mic normal (default.) If I record a call, the caller is way, way
quieter than the person using the microphone. Why is this?
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 2:26 PM, Norman Franke wrote:
As far as I can see, it may be when there are, say 3 sources, in
the conference being mixed, then we drop down to two. So if the
"level" is, say 1, we were dividing by 3 and now it's by 2.
Norman Franke
ASD, Inc.
On Jan 9, 2008, at 12:16 PM, Norman Franke wrote:
More on this one. In conference.c line 1797, if I comment out
this as follows:
// conf_port->last_level = level;
Then don't get a click when the playback finishes. Seems like
maybe there is an issue when the number of transmitting ports drops?
-Norman
On Jan 9, 2008, at 11:35 AM, Norman Franke wrote:
On Jan 8, 2008, at 5:03 PM, Norman Franke wrote:
This may not be related, but I've noticed that if I zero the mic
volume and play a file, it sounds great. If I keep the volume at
1.0 and play file, I get a loud pop after the file finishes.
It's
random, so it doesn't always do it, but seems to more often
than not.
This may have more to do with adding and removing sources from the
bridge. Still getting loud clicks and pops.
-Norman
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org
Visit our blog: http://blog.pjsip.org
pjsip mailing list
pjsip@lists.pjsip.org
http://lists.pjsip.org/mailman/listinfo/pjsip_lists.pjsip.org