[vls-devel] Re: Occasional crazy problem with vls

[Andrew de Quincey]

On Monday 19 January 2004 15:11, Jean-Paul Saman wrote:
> Andrew de Quincey wrote:
> > On Monday 19 January 2004 14:47, Jean-Paul Saman wrote:
> >>Andrew de Quincey wrote:
> >>>>But did your clients expirience a still frame or black frame for some
> >>>>time when copying the large file ? you haven't answered this question.
> >>>
> >>>No. On VLC, the first thing I see wrong is the speeded up play. No
> >>> stills or black frames.
> >>>
> >>>On our STBs, which are (unfortunately) far less tolerant to problems,
> >>> all kinds of crap happens... jitter, mpeg blocking, black frames...
> >>
> >>Hmm, sounds like another problem we solved earlier with VLS. It turned
> >>out that with *weird* PCR values strange things happen (Weird in the
> >>sense of a broken MPEG clock timeline). It should be solved in the CVS,
> >>unfortunately I do not know which branch has that fix and which doesn't.
> >
> > I'm pretty certain the file's PCR values are fine. It streams perfectly
> > under normal system load situations, and none of the tools I have run it
> > through report any weird PCRs or other problems.
>
> Then I go back to my first analysis of the problem. The copy starves the
> VLS reads.
>
> Does your system have DMA enabled on the HD?

Yes. Its a Maxtor ATLASU320_18_SCA SCSI disk.

> Which kernel are you running?

2.4.21

I think basically the problem is we didn't choose those servers to have 
particularly great disk IO throughput.. this streaming of one file is 
basically just a feature to keep a particular client happy. We use them to 
read the data to be streamed directly from DVB or MPEX cards and then send it 
to the network interface. 

In the future, we would properly spec streaming machines etc.. but in the 
meantime. sigh.

Mind you, I would have thought it would be able to cope with streaming one 
5mbit/s file. 

Anyway, the problem only occurs infrequently. We're not in the habit of 
uploading huge files; basically once the system is going, it gets left alone. 
My patch seems to fix the runaway problem if it occurs, so I'm quite happy 
with it as it is now. I'll tidy up the patch and commit it tomorrow.

Could this problem affect our DVB streaming as well, but we've never noticed 
it? Because the speed at which DVB data can be retransmitted by VLS is 
limited by the speed at which it is received by the DVB hardware, it doesn't 
matter that VLS's timing goes off.

-- 
This is the vls-devel mailing-list, see http://www.videolan.org/streaming/
To unsubscribe, please read http://developers.videolan.org/lists.html
If you are in trouble, please contact <postmaster at videolan.org>