Hi all,

Just a little addendum on 'catchups' that we were experiencing (and  
finally solved):

Our IP uses FreeBSD servers, and our particular server has massive  
resources,
lots of CPU power, 2.5 GB RAM, and abundant Bandwidth - 5 TB. So, tracking  
down these
devils was a real nightmare.

What gave use the clue to our problem was the slight differences in actual  
time that
the server displayed. We noticed that it slowly began to drift from a few  
ms to a few minutes. When we finally
looked into it, we found that FreeBSD resyncs server-time from an external  
source using ntpd.

Essentially, the server OS had to be re-compiled properly to allow users  
to update for time-drift.

Once this issue was attended to, the catchups have all but disappeared --  
finally.

A rather obtuse problem/solution. However, it might be worth checking for  
actual time-drift, and eliminating that posibility since this will affect  
liquidsoap performance.

Hope this might help someone out there in LS Land :)
kronos



On Tue, 18 Jun 2013 23:17:38 -0400, David Baelde <[email protected]>  
wrote:

> Hi guys,
>
> These issues are hard to fix, as they depend on a lot of factors. We
> should keep watching out for what liquidsoap could do better, but so
> far I don't see how the tool itself could help here.
>
> Instead, I thought I would do a quick recap on what overruns/underruns
> and "we need to catchup" are and what causes them. You may know all
> this already, and it might not help much because in the end anything
> can be caused by anything. In a sense, all it says is that liquidsoap
> can't do any better on that... please contradict me on that :p
>
> The catchup thing is when liquidsoap did not manage to compute its
> stream fast enough for sending on time. In my experience this is
> caused by lack of CPU, so I'd advise to look this way. It could be
> that a latency on the filesystem or network is making things late, but
> we are trying to minimize this kind of interference because streaming
> on time is the priority. The bottom line is look for CPU first, but if
> you think you've spotted another cause let us know so we can discuss
> potential solutions.
>
> A buffer underrun could be caused by a network lag, a slow source, or
> liquidsoap being too slow on something else to read from the source
> and fill its buffer. A buffer overrun could also be caused by
> liquidsoap being too slow to consume the buffer. (A typical situation
> where you get lots of overruns is if you don't always read a live
> stream, the solution is to put a dummy output on that stream so that
> it always gets consumed.) However, it could also be that the source is
> too fast. This happens: not all computers have the same clock, and not
> all streaming software deal with time in the same way. If you have
> relatively rare and periodic overruns that don't seem to be correlated
> with CPU spikes or catchup, this is probably the cause.
>
> Cheers,
>
> David
>
> On Mon, Jun 17, 2013 at 8:53 AM, Akos Veres <[email protected]> wrote:
>> I'm experiencing the same issues every day, didn't quite manage to fix  
>> it. I
>> had bigger drouputs, like streams actually stopping, then I stopped the
>> skip_blank feature thinking my CPU is under heavy dosage, it did make  
>> the
>> stream drops stop, but the overrun is still there.
>>
>> Any fix is welcome.
>>
>> Best regards,
>>         Ákos Veres
>>
>> _____________________________________________
>> http://akos.me - A little about me.
>>
>>
>> On Fri, Jun 14, 2013 at 6:15 PM, Alexander Schremmer  
>> <[email protected]>
>> wrote:
>>>
>>> On 12.06.2013 10:20, Alexander Schremmer wrote:
>>>
>>> >>> Are the times when this problem happens consistent? Try setting the
>>> >>> the
>>> >>> CPU priority of Liquidsoap to the max and see if that helps:
>>> >>
>>> >> No, but the other processes might demand resources irregularly.
>>> >>
>>> >> I will give it a boost to -5 (-19 sounds a bit hazardous given that  
>>> I
>>> >> have seen LS eating 100% in the past).
>>> >
>>> > Now I have another issue left: after I set a metadata item via  
>>> telnet, I
>>> > get spurious skips that look like this in the logs:
>>> >
>>> >> 2013/06/12 10:16:24 [server:3] New client: localhost.
>>> >> 2013/06/12 10:16:24 [server:3] Client localhost disconnected without
>>> >> saying goodbye..!
>>> >> 2013/06/12 10:16:26 [clock.wallclock_main:2] We must catchup 1.05
>>> >> seconds!
>>> >
>>> > And the audio skips for maybe 300 ms in the stream when listening.  
>>> What
>>> > could be the reason? Interestingly, it always happens shortly after
>>> > these telnet connections but never in a song.
>>>
>>> And today we got 90 seconds outage because of a cascade of problems in
>>> Liquidsoap. It started with the error message "Buffer overrun: Dropping
>>> 0.03s." and ended with both icecast sinks reconnecting, and then it
>>> needed still 80 seconds till they started working.
>>> Anybody an idea what could have caused this?
>>>
>>> Kind regards,
>>> Alexander
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> This SF.net email is sponsored by Windows:
>>>
>>> Build for Windows Store.
>>>
>>> http://p.sf.net/sfu/windows-dev2dev
>>> _______________________________________________
>>> Savonet-users mailing list
>>> [email protected]
>>> https://lists.sourceforge.net/lists/listinfo/savonet-users
>>
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Windows:
>>
>> Build for Windows Store.
>>
>> http://p.sf.net/sfu/windows-dev2dev
>> _______________________________________________
>> Savonet-users mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/savonet-users
>>
>
>
>


-- 
Using Opera's revolutionary email client: http://www.opera.com/mail/

------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:

Build for Windows Store.

http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Savonet-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/savonet-users

Reply via email to