As you've noted/realized, any packet loss - other than network packet
loss - that occurs in a system that uses our code *must* be the fault
of insufficient buffering inside the OS. I.e., this is an OS
problem, and any fix/workaround needs to address this.
Unfortunately there is no good way for our code to handle a failed
network write (i.e., no good way to 'retry' the write later).
Instead, we'll need to set things up to ensure (with high
probability) that network writes do not fail. Note also that we
can't try to avoid these write failures by making network sockets
blocking. The calls to "makeSocketNonBlocking()" are there for a
good reason (e.g., to overcome the problem that Marc Neuberger noted
in
<http://lists.live555.com/pipermail/live-devel/2007-June/006890.html>),
so removing those calls is *not* an acceptable solution.
The simplest solution, it seems, is simply to make sure that there is
reasonable buffering inside the OS on writeable sockets. To do this,
add
increaseSendBufferTo(envir(), clientSocket, 50*1024);
after the call to "makeSocketNonBlocking()" on line 235 of
"liveMedia/RTSPServer.cpp".
Similarly, add
increaseSendBufferTo(envir(), fGS->socketNum(), 50*1024);
after the call to "makeSocketNonBlocking()" on line 101 of
"liveMedia/RTPInterface.cpp".
These additions will be made to the next release of the code.
--
Ross Finlayson
Live Networks, Inc.
http://www.live555.com/
_______________________________________________
live-devel mailing list
live-devel@lists.live555.com
http://lists.live555.com/mailman/listinfo/live-devel
