The infinite loops role is to send out a keep alive signal on the socket to all the clients every X seconds.

One simple mechanism would be to have the child process (or thread) connect back to the listening socket and send a message every X seconds. Your existing code without modification would forward these to each of the clients as any other message. Though you might want to avoid sending stuff to the keep-alive client. Once you start discriminating about what messages you forward to what clients, the keep-alive just gets forward to all clients (except it's originator) as now.

Of course, the next step is likely to be to only send keep-alive messages to those clients that you haven't sent any message to (and perhaps, received a message from) for the last X seconds. It's at this point that the thread would come into it's own, as you can then use a shared hash to record the last time a message was sent to each client.

The keep-alive thread then wakes up every X/n (where n depends upon the resolution you require), scans the shared hash, and sends keep-alive messages, with a client identifier parameter, for each client who's time falls due within the next X/n period. Your main select loop will receive and forward these messages in the normal way.