Stuttering performance?

[MVPMC-2]

Hi!

First, this is a great app, but I'm having some problems thatm aybe
someone can help me with. I'm running the current build of the mvpmc code.

Generally they both work just fine. But after a random period of time
of use (can be a half hour, can be a few hours, but more the former
than the latter), I get stuttering of the video/audio. I have two
mvpmc devices, one wireless, one not and it's true for both of them
(even when one is running when the other one is not).

When I look at the  on screen display the demux info shows that the
video (and audio) buffer is empty. If I pause the mvpmc, the buffer
fills back up and it works for a few seconds, and then back to stuttering.

Most of the time a complete power down of the device and reboot
appears to fix it at least for a while.

The service it's pulling data from is effectively without load. I
used to get some skipping whenever mythtv started a new recording,
but an update of the mvpmc fixed it (I saw that a fix had been
posted). This appears unrelated to that, as the problem occurs even
when there's nothing scheduled in myth. watching directly from myth
appears to be ok.

the server is running at 1 GB, and it's running the WD "Green"
drives. While they're not supposed to be the greatest performers in
the world, they should be able to handle the streaming of 5 mb/sec or
so that the demux for the mvpmc shows. Their max is 3 Gb/s, and
that's almost 1000 times faster than I'm looking for.  the CPU never
shows any load (I don't think I've ever seen the load over about   1 under top)

the only other thing on the system is zoneminder and a video capture
card. I've wondered if that was it (resource or performance
conflict), but the problem appears even with zoneminder disabled.

Any ideas? Heat, buffering, and memory leaks are all I can think of.

Thanks

Rick



Rick Steeves
http://www.sinister.net

"The journey is the destination"


------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

MVPMC@... wrote:

> ...after a random period of time
> of use (can be a half hour, can be a few hours, but more the former
> than the latter), I get stuttering of the video/audio.
>
> When I look at the  on screen display the demux info shows that the
> video (and audio) buffer is empty. If I pause the mvpmc, the buffer
> fills back up and it works for a few seconds, and then back to stuttering.
>
> Most of the time a complete power down of the device and reboot
> appears to fix it at least for a while.
>
> Any ideas? Heat, buffering, and memory leaks are all I can think of.

It's the mvpmc running out of memory, due to memory leaks and
inefficiencies in how the program data is stored. (See discussion on the
dev list.)

You can work around the problem by pruning your show database in MythTV.
For example, if there's stuff you're archiving, export it with
nuvexport, and then delete it in MythTV.

Or just get used to having to do a warn restart periodically. I think
you'll find that doing a full reboot is not necessary.

I used to see this symptom quite regularly, but since two things
happened, I run out of memory so often that I pretty much need to do a
warn restart after viewing any video. (One of the things that happened
is that I had to downgrade mvpmc to a version prior to the JIT audio
changes, because it messed up sync for me (as documented in an earlier
posting), and the older version has more memory problems. The other
thing is that I expanded my MythTV storage, so my program database is
even bigger.)

  -Tom

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 09:30 PM 3/2/2009, Tom Metro wrote:

>It's the mvpmc running out of memory, due to memory leaks and
>inefficiencies in how the program data is stored. (See discussion on
>the dev list.)


Are there are buffer settings that might improve things?

Rick



------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 10:23 PM 3/2/2009, MVPMC@... wrote:
>Are there are buffer settings that might improve things?

 From the threads I read on the devel list, it looks like the problem
is for 1500 ish recordings? I've only got a few hundred .... *sigh*.

Rick

>Rick
>
>
>
>------------------------------------------------------------------------------
>Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
>-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
>-Strategies to boost innovation and cut costs with open source participation
>-Receive a $600 discount off the registration fee with the source code: SFAD
>http://p.sf.net/sfu/XcvMzF8H
>_______________________________________________
>Mvpmc-users mailing list
>Mvpmc-users@...
>https://lists.sourceforge.net/lists/listinfo/mvpmc-users
>mvpmc wiki: http://mvpmc.wikispaces.com/

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

MVPMC@... wrote:
> Tom Metro wrote:
>> It's the mvpmc running out of memory, due to memory leaks and
>> inefficiencies in how the program data is stored. (See discussion on
>> the dev list.)
>
> Are there are buffer settings that might improve things?

You mean boosting the size of the audio/video buffers to ride out the
slow response of the mvpmc? No. Unfortunately once the mvpmc starts
hitting the memory wall, it slows down continuously from that point
forward, so increasing buffers will only delay the inevitable, and take
up more memory.


> From the threads I read on the devel list, it looks like the problem
> is for 1500 ish recordings? I've only got a few hundred .... *sigh*.

Part of the problem is memory leaks, which as you'd expect get
progressively worse over time. I first started seeing symptoms when I
had in the hundreds of recordings. More recordings just make the
symptoms noticeable sooner.

It shouldn't be a substantial problem to fix. Roger Heflin posted
results of research he did, which I think identified some memory leaks
that could be plugged. I proposed some architecture tweaks that would
filter program data based on the selected recording groups, which if you
use recording groups, could substantially cut down what needs to be
loaded into memory. It all just waiting for the right person with
adequate time, motivation, and skills to implement it...

  -Tom

------------------------------------------------------------------------------
Open Source Business Conference (OSBC), March 24-25, 2009, San Francisco, CA
-OSBC tackles the biggest issue in open source: Open Sourcing the Enterprise
-Strategies to boost innovation and cut costs with open source participation
-Receive a $600 discount off the registration fee with the source code: SFAD
http://p.sf.net/sfu/XcvMzF8H
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 04:45 PM 3/3/2009, Tom Metro wrote:
>MVPMC@... wrote:
>>Tom Metro wrote:
>>>It's the mvpmc running out of memory, due to memory leaks and
>>>inefficiencies in how the program data is stored. (See discussion
>>>on the dev list.)
>>Are there are buffer settings that might improve things?


I thought I"d follow up to say that I'm still seeing this. what's
most interesting is that a warm reboot of the device doesn't fix it
once it starts happening.  A cold boot also doesn't always appear to
fix it. (sometimes yes, sometimes no). Whe it is working it can work
for hours. once it starts happening tho the whole system becomes
unstable. Makes me wonder if there's a temp issue on the PVR

I have seen a soft reboot of the mythtv server appear to fix it, but
I don't have enough data there yet.





------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 01:13 AM 3/16/2009, mvpmc@... wrote:
>I have seen a soft reboot of the mythtv server appear to fix it, but
>I don't have enough data there yet.


I've now fixed the stuttering twice by leaving the mvpmc powered on
when it was stuttering, rebooting the mythtv server, and then
restarting the show that was stuttering.

Things play fine directly from the mythtv.  I don't see any mysql or
mythtv log errors.

Rick


------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

mvpmc@... wrote:
> I've now fixed the stuttering twice by leaving the mvpmc powered on
> when it was stuttering, rebooting the mythtv server, and then
> restarting the show that was stuttering.

So with some limited data, the trend seems to be indicating that it is a
back-end problem...


> Things play fine directly from the mythtv.  I don't see any mysql or
> mythtv log errors.

...and you're wondering why it is impacting only the mvpmc client?

Is your other MythTV client remote? Is it running over the same network
(segments)?

Have you looked at how loaded your network is during the incidence of
stuttering?

Have you tried using the bandwidth test feature of mvpmc both during and
after the stuttering symptoms? How about using the bandwidth test while
browsing the file system (I think that's possible). Assuming you have at
least one network file system mounted to your MVP, that might help
narrow down whether it is a network issue, or a specific issue with the
MythTV back-end server.

  -Tom


------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 10:28 PM 3/16/2009, Tom Metro wrote:
>mvpmc@... wrote:
>So with some limited data, the trend seems to be indicating that it
>is a back-end problem...
>
>...and you're wondering why it is impacting only the mvpmc client?

yup.

>Is your other MythTV client remote? Is it running over the same
>network (segments)?

My network is moderately boring. there's little traffic on it, and
the backbone (which includes the myth server) is GB.  My only other
way to view Mythtv is on the frontend on the server.

>Have you looked at how loaded your network is during the incidence
>of stuttering?

yup, and the answer has been, effectively, nada.  I have backups that
run at 2 am. I have only one backend, so myth records locally (and I
haven't seen a correlation between myth recording and stuttering).

>Have you tried using the bandwidth test feature of mvpmc both during
>and after the stuttering symptoms?

ahh, see this is why I ask here! bandwidth test
feature?  *rummage*  well in a non-stuttering state I see 157 mb/s;
I'll follow up when I get failure data - THANKS.

also occurs to me from your comments I could try to view files across
the file system when it's stuttering and see if the problem is
independent of myth - also helpful.

>How about using the bandwidth test while browsing the file system (I
>think that's possible). Assuming you have at least one network file
>system mounted to your MVP, that might help narrow down whether it
>is a network issue, or a specific issue with the MythTV back-end server.

I do have a network file system mounted, but it's not being used for
anything at the time of stuttering; I'm only using the mythtv
interface. I do have two mvpmc systems (one wired, one wireless), but
the wireless one is not in use/usually off (and hasn't been on when
there was stuttering.)  The one I'm using is the wired one.

thx!

Rick


>  -Tom




------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Scott Souter]

I would like to comment that I have these exact same problems with my two
mvpmc's. I also have a frontend on my backend server that experiences no
issue. I recently added another frontend on my Apple iMac and it also has
no issue. The problems seem to be a problem with the mvpmc's only. If I
press pause on  for a few seconds, things seem to straighten out for a
while.  It seems like a buffer limitation or something.

Scott

> At 10:28 PM 3/16/2009, Tom Metro wrote:
>>mvpmc@... wrote:
>>So with some limited data, the trend seems to be indicating that it
>>is a back-end problem...
>>
>>...and you're wondering why it is impacting only the mvpmc client?
>
> yup.
>
>>Is your other MythTV client remote? Is it running over the same
>>network (segments)?
>
> My network is moderately boring. there's little traffic on it, and
> the backbone (which includes the myth server) is GB.  My only other
> way to view Mythtv is on the frontend on the server.
>
>>Have you looked at how loaded your network is during the incidence
>>of stuttering?
>
> yup, and the answer has been, effectively, nada.  I have backups that
> run at 2 am. I have only one backend, so myth records locally (and I
> haven't seen a correlation between myth recording and stuttering).
>
>>Have you tried using the bandwidth test feature of mvpmc both during
>>and after the stuttering symptoms?
>
> ahh, see this is why I ask here! bandwidth test
> feature?  *rummage*  well in a non-stuttering state I see 157 mb/s;
> I'll follow up when I get failure data - THANKS.
>
> also occurs to me from your comments I could try to view files across
> the file system when it's stuttering and see if the problem is
> independent of myth - also helpful.
>
>>How about using the bandwidth test while browsing the file system (I
>>think that's possible). Assuming you have at least one network file
>>system mounted to your MVP, that might help narrow down whether it
>>is a network issue, or a specific issue with the MythTV back-end server.
>
> I do have a network file system mounted, but it's not being used for
> anything at the time of stuttering; I'm only using the mythtv
> interface. I do have two mvpmc systems (one wired, one wireless), but
> the wireless one is not in use/usually off (and hasn't been on when
> there was stuttering.)  The one I'm using is the wired one.
>
> thx!
>
> Rick
>
>
>>  -Tom
>
>
>
>
> ------------------------------------------------------------------------------
> Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
> powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
> easily build your RIAs with Flex Builder, the Eclipse(TM)based development
> software that enables intelligent coding and step-through debugging.
> Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
> _______________________________________________
> Mvpmc-users mailing list
> Mvpmc-users@...
> https://lists.sourceforge.net/lists/listinfo/mvpmc-users
> mvpmc wiki: http://mvpmc.wikispaces.com/
>

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 12:30 PM 3/20/2009, scott@... wrote:
>It seems like a buffer limitation or something.


In my case when looking at the buffer data on the mvpmc the buffer is
definitely empty. If I hit pause I can see the buffer refill, but
when I unpause the mypmc buffer almost immediately drains back to
zero. So I can watch for 8-10 sec and then back to stuttering.

Rebooting the backend server fixes the problem for 30 min to about 3 hours.

Rick


------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Jamin W. Collins]

mvpmc@... wrote:

> At 12:30 PM 3/20/2009, scott@... wrote:
>> It seems like a buffer limitation or something.
>
>
> In my case when looking at the buffer data on the mvpmc the buffer is
> definitely empty. If I hit pause I can see the buffer refill, but
> when I unpause the mypmc buffer almost immediately drains back to
> zero. So I can watch for 8-10 sec and then back to stuttering.
>
> Rebooting the backend server fixes the problem for 30 min to about 3 hours.

I had a problem similar to this, every once in a while the playback
through mvpmc would stutter or pause for a while.  I eventually tracked
the problem back to iowait due to failing drives.  Essentially, the
drives couldn't keep up with all that was being requested of them.  A
few days later the drives actually failed.  Since replacing the bad
drives, performance has been smooth.

--
Jamin W. Collins

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

mvpmc@... wrote:
> In my case when looking at the buffer data on the mvpmc the buffer is
> definitely empty. If I hit pause I can see the buffer refill, but
> when I unpause the mypmc buffer almost immediately drains back to
> zero. So I can watch for 8-10 sec and then back to stuttering.

This exactly describes a behavior I observe on rare occasion, but...


> Rebooting the backend server fixes the problem for 30 min to about 3 hours.

In my case merely stopping and restarting the video resolves the
problem. Unless mvpmc is also showing signs of hitting its memory
limits, in which case a warm restart does it. I've never had to restart
my back-end to resolve an mvpmc issue.

On the other hand, until I tweaked my back-end to run its monthly RAID
rebuilds at a lower priority, I'd see persistent "starving buffer"
problems for the duration of the rebuild. As another poster mentioned,
poor disk I/O performance can lead to these symptoms.

  -Tom


------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[MVPMC-2]

At 10:23 PM 3/20/2009, Tom Metro wrote:
>On the other hand, until I tweaked my back-end to run its monthly
>RAID rebuilds at a lower priority, I'd see persistent "starving
>buffer" problems for the duration of the rebuild. As another poster
>mentioned, poor disk I/O performance can lead to these symptoms.

Given that it runs for several hours sometimes with no issues it
would seem odd it would be a disk IO problem. The buffer pretty much
just sits there full until I start to see problems, and no amount of
stopping the video resolves the problem. Pause will refill the
buffer,  but once unpaused it almost instantly empties. Once it's
stuttering it doesn't recover. (at least over intervals of a half
hour or so, which should clearly resolve a buffering problem.

The system is WD Green Caviar drives, non-RAIDed. One single drive is
used exclusively for Myth. I get no SMART issues, or errors in
/var/log/messages.

If you have pointers for debug beyond what I've posted so far I'd
appreciate any insights.

Rick


>  -Tom




------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Roger Heflin-3]

mvpmc@... wrote:

> At 10:23 PM 3/20/2009, Tom Metro wrote:
>> On the other hand, until I tweaked my back-end to run its monthly
>> RAID rebuilds at a lower priority, I'd see persistent "starving
>> buffer" problems for the duration of the rebuild. As another poster
>> mentioned, poor disk I/O performance can lead to these symptoms.
>
> Given that it runs for several hours sometimes with no issues it
> would seem odd it would be a disk IO problem. The buffer pretty much
> just sits there full until I start to see problems, and no amount of
> stopping the video resolves the problem. Pause will refill the
> buffer,  but once unpaused it almost instantly empties. Once it's
> stuttering it doesn't recover. (at least over intervals of a half
> hour or so, which should clearly resolve a buffering problem.
>
> The system is WD Green Caviar drives, non-RAIDed. One single drive is
> used exclusively for Myth. I get no SMART issues, or errors in
> /var/log/messages.
>
> If you have pointers for debug beyond what I've posted so far I'd
> appreciate any insights.
>
> Rick
>
>

If something were running on the drive (a cron job, a big copy) that
would explain the behavior.   And rebooting would "FIX" it since it
would kill the job.

On mine I have some jobs that copy around large files a couple of
times a day, those jobs did not cause stuttering...they completely
stopped the mvpmc until the job finished.   I finally went in and used
   rsync with a -bwlimit option to limit how hard it hits the disk.

Since the mvpmc had very small buffers it cannot tolerate it taking
very long for data to get to it, so is very sensitive to io delay on
the backend.

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Jamin W. Collins]

mvpmc@... wrote:

> Given that it runs for several hours sometimes with no issues it
> would seem odd it would be a disk IO problem. The buffer pretty much
> just sits there full until I start to see problems, and no amount of
> stopping the video resolves the problem. Pause will refill the
> buffer,  but once unpaused it almost instantly empties. Once it's
> stuttering it doesn't recover. (at least over intervals of a half
> hour or so, which should clearly resolve a buffering problem.
>
> The system is WD Green Caviar drives, non-RAIDed. One single drive is
> used exclusively for Myth. I get no SMART issues, or errors in
> /var/log/messages.
>
> If you have pointers for debug beyond what I've posted so far I'd
> appreciate any insights.

When the mvpmc exhibits the problem, check the iowait percentage on the
myth backend.  Using top should work.  The iowait percentage should be
low to non-existent on a healthy system.

--
Jamin W. Collins

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

mvpmc@... wrote:
> Given that it runs for several hours sometimes with no issues it would
> seem odd it would be a disk IO problem.

It would be perfectly logical if the I/O overload is caused by some
sporadic operation on the back-end.

For example, how many tuners do you have? If you have two, maybe the
slowdown only occurs when both are recording.

Or as Roger mentioned, it could be something unrelated to MythTV
happening on the server. That was also the point of my story about the
RAID rebuilds.


> If you have pointers for debug beyond what I've posted so far I'd
> appreciate any insights.

Looking at the iowait, as Jamin suggested, seems like a good idea.
There's an iostat program that might prove useful as well. But I
wouldn't overlook CPU load either. The problem I ran into with RAID
rebuilds was originally thought to be an I/O problem, but ended up being
CPU.


You previously posted:
> bandwidth test feature? *rummage* well in a non-stuttering state I
> see 157 mb/s; I'll follow up when I get failure data...

Have you had an opportunity to run the throughput test while
experiencing the symptoms?

157 mb/s seems exceptionally high. I see 9 mb/s via MythTV, and 15 mb/s
via SMB. I think 8 to 20 mb/s is the typical range. Run some searches on
the list archives to see what others have reported.

  -Tom

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Scott Souter]

> mvpmc@... wrote:
>> In my case when looking at the buffer data on the mvpmc the buffer is
>> definitely empty. If I hit pause I can see the buffer refill, but
>> when I unpause the mypmc buffer almost immediately drains back to
>> zero. So I can watch for 8-10 sec and then back to stuttering.
>
> This exactly describes a behavior I observe on rare occasion, but...
>
>
>> Rebooting the backend server fixes the problem for 30 min to about 3
>> hours.
>
> In my case merely stopping and restarting the video resolves the
> problem. Unless mvpmc is also showing signs of hitting its memory
> limits, in which case a warm restart does it. I've never had to restart
> my back-end to resolve an mvpmc issue.
>
> On the other hand, until I tweaked my back-end to run its monthly RAID
> rebuilds at a lower priority, I'd see persistent "starving buffer"
> problems for the duration of the rebuild. As another poster mentioned,
> poor disk I/O performance can lead to these symptoms.
>

I see this same problem.  I am started to wonder now if my problem is
related to my RAID setup.  I use 4 disks in a software RAID 5 array.  How
would one sort out or even find out if this could be the source of a
problem?  In my case this backend box is used only for Myth and the odd
Googling about from time to time.

Thx
Scott

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Tom Metro-10]

scott@... wrote:
> I use 4 disks in a software RAID 5 array.  How
> would one sort out or even find out if this could be the source of a
> problem?

It's pretty easy. Wait until the symptoms occur (by default the mdadm
package on Ubuntu schedules a rebuild starting early in the AM on the
first Sunday of the month). Then launch the system monitor from the GUI
or use the equivalent command line tools (ps or top) to find a process
with a name like md0_resync and drop its priority to something low, like
-15. If that resolves the symptom, then you should install a permanent fix.

On my system I modified /etc/cron.d/mdadm to run a few additional
commands to drop the priority. Onto the end of the existing command
line, I tacked:
  && sleep 60 && renice 15 `pidof -s -x md1_resync` > /dev/null

and that permanently solved it for me.

Prior to this I spent months playing around with the MD driver parameter
speed_limit_min, both manually via /proc, and via /etc/sysctl.conf
(dev.raid.speed_limit_min), but eventually determined that the system
was CPU bound rather than I/O bound during rebuilds.

  -Tom

------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/

[Scott Souter]

> scott@... wrote:
>> I use 4 disks in a software RAID 5 array.  How
>> would one sort out or even find out if this could be the source of a
>> problem?
>
> It's pretty easy. Wait until the symptoms occur (by default the mdadm
> package on Ubuntu schedules a rebuild starting early in the AM on the
> first Sunday of the month). Then launch the system monitor from the GUI
> or use the equivalent command line tools (ps or top) to find a process
> with a name like md0_resync and drop its priority to something low, like
> -15. If that resolves the symptom, then you should install a permanent
> fix.
>
> On my system I modified /etc/cron.d/mdadm to run a few additional
> commands to drop the priority. Onto the end of the existing command
> line, I tacked:
>   && sleep 60 && renice 15 `pidof -s -x md1_resync` > /dev/null
>
> and that permanently solved it for me.
>
> Prior to this I spent months playing around with the MD driver parameter
> speed_limit_min, both manually via /proc, and via /etc/sysctl.conf
> (dev.raid.speed_limit_min), but eventually determined that the system
> was CPU bound rather than I/O bound during rebuilds.
>

Hmmm maybe not so easy in my case.  My Asus MB uses an onboard Nvidia
"software RAID" that is controlled at the bios level.  The only
involvement for the 'nix OS is to ensure that the kernel can see the array
properly (which was no picnic).

Any ideas on that front?  I am not sure how one even determines if it's an
issue.

Scott


------------------------------------------------------------------------------
Apps built with the Adobe(R) Flex(R) framework and Flex Builder(TM) are
powering Web 2.0 with engaging, cross-platform capabilities. Quickly and
easily build your RIAs with Flex Builder, the Eclipse(TM)based development
software that enables intelligent coding and step-through debugging.
Download the free 60 day trial. http://p.sf.net/sfu/www-adobe-com
_______________________________________________
Mvpmc-users mailing list
Mvpmc-users@...
https://lists.sourceforge.net/lists/listinfo/mvpmc-users
mvpmc wiki: http://mvpmc.wikispaces.com/