通过在 groupsock\GroupsockHelper.cpp 中打断点，发现一个rtsp over tcp的客户端连接上来，共调用了increaseSendBufferTo函数6次。打印出来的每次调用的socket号和对应的buffer size分别为
increaseBufferTo socket=824 size=51200
increaseBufferTo socket=972 size=51200
increaseBufferTo socket=976 size=51200
increaseBufferTo socket=988 size=51200
increaseBufferTo socket=988 size=1125000
increaseBufferTo socket=992 size=51200
而最终 RTPInterface::sendDataOverTCP 使用的socket为972的那个，就是第二个调用increaseSendBufferTo的那个socket。这意味着，tcp的发送缓冲只有51200字节。下面再具体分析上面这6次increaseSendBufferTo分别发生在什么地方
第一次调用，发生在 RTSPServer::createNew，这个函数接着调用了GenericMediaServer::setUpOurSocket
ourSocket = setupStreamSocket(env, ourPort);
if (ourSocket < 0) break;
// Make sure we have a big send buffer:
if (!increaseSendBufferTo(env, ourSocket, 50*1024)) break;
// Allow multiple simultaneous connections:
if (listen(ourSocket, LISTEN_BACKLOG_SIZE) < 0) {
env.setResultErrMsg("listen() failed: ");
break;
}
可以看出，这是针对侦听的socket进行操作，意义不大
第二次调用，发生在 GenericMediaServer::incomingConnectionHandlerOnSocket
struct sockaddr_in clientAddr;
SOCKLEN_T clientAddrLen = sizeof clientAddr;
int clientSocket = accept(serverSocket, (struct sockaddr*)&clientAddr, &clientAddrLen);
if (clientSocket < 0) {
int err = envir().getErrno();
if (err != EWOULDBLOCK) {
envir().setResultErrMsg("accept() failed: ");
}
return;
}
ignoreSigPipeOnSocket(clientSocket); // so that clients on the same host that are killed don't also kill us
makeSocketNonBlocking(clientSocket);
increaseSendBufferTo(envir(), clientSocket, 50*1024);
这是针对连接进来的tcp socket，也就是之后进行命令交互和rtp数据交互的tcp通道，进行的设置。
第三次调用，发生在 RTSPServer::RTSPClientConnection
::handleCmd_DESCRIBE。这个函数中会调用 ServerMediaSession::generateSDPDescription，从而最终创建了一个RTPInterface对象。
RTPInterface::RTPInterface(Medium* owner, Groupsock* gs)
: fOwner(owner), fGS(gs),
fTCPStreams(NULL),
fNextTCPReadSize(0), fNextTCPReadStreamSocketNum(-1),
fNextTCPReadStreamChannelId(0xFF), fReadHandlerProc(NULL),
fAuxReadHandlerFunc(NULL), fAuxReadHandlerClientData(NULL) {
// Make the socket non-blocking, even though it will be read from only asynchronously, when packets arrive.
// The reason for this is that, in some OSs, reads on a blocking socket can (allegedly) sometimes block,
// even if the socket was previously reported (e.g., by "select()") as having data available.
// (This can supposedly happen if the UDP checksum fails, for example.)
makeSocketNonBlocking(fGS->socketNum());
increaseSendBufferTo(envir(), fGS->socketNum(), 50*1024);
}
这里设置的是RTPInterface对应的Groupsock，组播的socket
第四、第五次调用，发生在 RTSPServer::RTSPClientSession
::handleCmd_SETUP。这里调用了 OnDemandServerMediaSubsession
::getStreamParameters
rtpSink = createNewRTPSink(rtpGroupsock, rtpPayloadType, mediaSource);
if (rtpSink != NULL && rtpSink->estimatedBitrate() > 0) streamBitrate = rtpSink->estimatedBitrate();
}
// Turn off the destinations for each groupsock. They'll get set later
// (unless TCP is used instead):
if (rtpGroupsock != NULL) rtpGroupsock->removeAllDestinations();
if (rtcpGroupsock != NULL) rtcpGroupsock->removeAllDestinations();
if (rtpGroupsock != NULL) {
// Try to use a big send buffer for RTP - at least 0.1 second of
// specified bandwidth and at least 50 KB
unsigned rtpBufSize = streamBitrate * 25 / 2; // 1 kbps * 0.1 s = 12.5 bytes
if (rtpBufSize < 50 * 1024) rtpBufSize = 50 * 1024;
increaseSendBufferTo(envir(), rtpGroupsock->socketNum(), rtpBufSize);
}
createNewRTPSink先创建了一个RTPInterface，其构造函数会设置一次发送缓冲大小。然后下面的代码会根据码率，对rtpGroupsock再设置一次缓冲大小。很明显，这两次调用increaseSendBufferTo都是针对udp的组播socket的。
第六次调用，发生在 RTSPServer::RTSPClientSession
::handleCmd_PLAY。这个函数调用了 OnDemandServerMediaSubsession::startStream，最终创建一个RTCPInstance对象。而RTCPInstance类中有一个RTPInterface fRTCPInterface的成员，因此又创建了一个RTPInterface，从而又调用了一次increaseSendBufferTo。
综上分析，可以看出rtsp的tcp通道只被设置了一次发送缓冲，其值为50 *1024。显然这个值太小了，对于发送大码率视频是不足够的