[Live-devel] scaling live555 rtsp proxy to hundreds of viewers.

[Jeff Shanab]

Are you making a copy for each connected viewer?
The system I worked on just over a year ago could stream around 400 streams
but never was it 400 of 1 stream, it was 5 or 10 of 100-200 sources.
Even then I used a buffer pool and a shared pointer so when the last
unicast client was sent the packet, it returned itself to the pool.

Another thing was my architecture, which may be important. I was RTSP in
from many sources and serving thru my own http protocol basterdization. In
this I ran "Processors" of a settable number of sources. Each Processor was
it's own usage environment and ran in a thread. This allowed the OS to
parallelize the I/O to the buffers to the ethernet interface.  IF the event
Queue design, whichdoes have  the best overall thruoughput, gets too deep,
(spread across 2 many sources) then latency can go up and slow overall
thruput to multiple clients.

I was able to get to the limit of the gigibit network.
I usually ran from 10 to 64 sources per "processor"
  If the cameras had small frames like D1 it was better to run 64 per
processor. (FD_SET on windows is slower beyond 64 sockets)
  If they were high resolution cameras and had big nal frames, it was
faster to break it into 6 processors of 10 and let the OS utilize the cores.
    (Threads tend to stay on a single core)

On Fri, Sep 26, 2014 at 4:40 PM, Pete Pulliam <ppulliam at llnw.com> wrote:

> I have an implementation of an rtsp proxy server based on the
> live555ProxyServer that ships with the live555 source.  When connecting
> only a couple of viewers to a single proxied stream, things look great.
> I'm hoping to get a few hundred viewers per stream though.  What I'm seeing
> is that there is a large drop off in quality and kbps as I add more
> viewers.  I see this drop off both with the custom proxy I've written and
> the stock proxy that ships with the source.
>
> While this is happening, the CPU and memory are only lightly used, and the
> NIC is not that busy.
>
> I'm trying to use this to shield a transcoder that provides an RTSP origin
> (as well as several other formats).  When I run the same test directly
> against the origin I don't see a drop off in quality like this for viewers
> less than 50 (and haven't tested that for more than 50 viewers).  Hitting
> the origin directly looks great.
>
> Is there perhaps something I should have tuned in the proxy, or could tune
> with the Linux box this is running on that is causing this drop off?
> Advice wanted to improve the scaling of single stream => many viewer.
>
> The kbps drop off looks like:
> # ​
> of clients
> ​ , ​
> bitrate (kbps)
> 2 977.688820
> 3 976.666311
> 4 936.494160
> 5 940.096328
> 6 944.486316
> 7 955.723431
> 8 945.044076
> 9 944.803396
> 10 930.465372
> 11 925.247045
> 12 931.066158
> 13 713.205068
> 14 486.331708
> 15 492.473134
> 16 485.362249
> 17 483.711453
> 18 490.636454
> 19 485.919289
> 20 487.862608
> 21 489.587224
>
> Pete
>
>
> The information in this message may be confidential.  It is intended
> solely for
> the addressee(s).  If you are not the intended recipient, any disclosure,
> copying or distribution of the message, or any action or omission taken by
> you
> in reliance on it, is prohibited and may be unlawful.  Please immediately
> contact the sender if you have received this message in error.
>
>
> _______________________________________________
> live-devel mailing list
> live-devel at lists.live555.com
> http://lists.live555.com/mailman/listinfo/live-devel
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.live555.com/pipermail/live-devel/attachments/20140926/71002419/attachment-0001.html>