[Live-devel] Jeremy Noring's problem with RTP-over-TCP

[Jeremy Noring]

On Thu, Nov 24, 2011 at 1:25 PM, Ross Finlayson <finlayson at live555.com>wrote:

> Receiving RTP over TCP in current releases with multiple sub-sessions is,
> by and large, broken as best I can tell due to something with the new
> synchronous API
>
>
> You mean "asynchronous API".
>
> We talked about your problem a lot on the mailing list back in July.  As
> far as I can tell, you're the only person having this problem with
> receiving RTP-over-TCP, so it may be inaccurate to describe it as being
> "broken".  (I also doubt that the asynchronous API is responsible.)  Back
> in July I wasn't able to explain the symptoms that you were seeing, but I
> still suspect that you might be using multiple threads incorrectly (by,
> perhaps inadvertently, accessing the same LIVE555 object(s) from multiple
> threads).  If you stop using multiple threads (as I've noted several times,
> you don't need multiple threads to concurrently receive multiple RTSP/RTP
> streams within a single LIVE555 application if you use the asynchronous
> API), then I suspect your problems will go away.
>
>
Yes, I meant asynchronous.  My apologies.

I am not using multiple threads; I've been using this library for almost
four years, so I'm well aware of the threading implications.  Every Live555
instance I work with is completely contained on a single thread, and the
only signaling that occurs is through watch variables.  Also, I could
easily see someone missing this issue because in the currently library the
behavior is masked.  You still get marginal performance due to the
workaround that was included in the library.  But I can still reproduce the
issue on demand, and I suspect other people A) aren't using TCP with
multiple subsessions or B) aren't looking closely.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.live555.com/pipermail/live-devel/attachments/20111124/66bac2c4/attachment-0001.html>