Does any one know why the current tcp stack only handles 4 successive client connections before failing ? Running this program, and connecting to the mbed via "telnet 192.168.10.231 80" work 4 times, and on the fifth attempt, just hangs indefinitely <<program /users/nkhorman/code/test_tcp_server/ >> TIA for any help on this.

client.send_all(str, sizeof(str)-1);

client.send_all(str, sizeof(str)-1, 3000);

// Avoid code like this in a thread:
while(1) {
    Thread::yield();
}

Hi Neal, thank you for your experiments on the new networking stack <<quote nkhorman>> Does any one know why the current tcp stack only handles 4 successive client connections before failing ? <</quote>> === Official Example === The <<user mbed_official>> TCP Server example does not have this problem, you can use it as a reference: <<program /users/mbed_official/code/TCPEchoServer/ >> === Socket API === In your example you are using a blocking call (default) to "##send_all##", if anything goes wrong in the connection with your client, your server will remain stuck there without any chance to recover: <<code>> client.send_all(str, sizeof(str)-1); <</code>> Try to change that with something like a 3 seconds timeout: <<code>> client.send_all(str, sizeof(str)-1, 3000); <</code>> === RTOS === Another thing you can do to increase the reliability of your program is to move the most stack intensive code to ##main##. As you can see from the [[/handbook/RTOS-Memory-Model|RTOS Memory Model]] documentation, the ##main## stack has room for growth in the whole free RAM memory. Any other thread has a fixed size stack that cannot grow at run time. Remember that RAM in a microcontroller is a very scarce resource. Each RTOS thread has a relevant cost in terms of RAM for its stack. You want to avoid having a thread staying idle: <<code>> // Avoid code like this in a thread: while(1) { Thread::yield(); } <</code>> In your example, you can reduce the total number of threads by one (for example, move the ##ThreadProcSocket## to ##main##). HTH, Emilio

client.set_blocking(false, 3000);
[...]
client.send_all(str, sizeof(str)-1);

Hi Neal, actually getting the blocking behaviour keeping the "##timeout=0##" default did look a bit counter-intuitive. I have changed that to a "##set_blocking(bool blocking, unsigned int timeout=1500)" call, therefore you can use in your server: <<code>> client.set_blocking(false, 3000); [...] client.send_all(str, sizeof(str)-1); <</code>> Cheers, Emilio