The first thing we want to look at is server.cpp. This is a pretty simple application that listens for datagrams at a known port and sends back a response. In order to implement a true "discovery" mechanism, the server will have to be a little bit more picky about who it responds to. We'll tackle that issue in the next tutorial though...
// $Id$ /* Our datagram server will, of course, need to create a datagram. We'll also need an address object so that we know where to listen. */ #include "ace/Log_Msg.h" #include "ace/SOCK_Dgram.h" #include "ace/INET_Addr.h" /* Use the typical TCP/IP port address for receiving datagrams. */ static const u_short PORT = ACE_DEFAULT_SERVER_PORT; int main (int, char**) { /* This is where we'll listen for datagrams coming from the clients. We'll give this address to the open() method below to enable the listener. */ ACE_INET_Addr local (PORT); /* A simply constructed datagram that we'll listen with. */ ACE_SOCK_Dgram dgram; /* Like most ACE objects, the datagram has to be opened before it can be uses. Of course, -1 on failure. A datagram will fail to open if there is already a datagram listening at the port we've chosen. It *is* OK to open a datagram at a port where there is an ACE_SOCK_Stream though. This is because datagrams are UDP and SOCK_Stream is TCP and the two don't cross paths. */ if (dgram.open (local) == -1) ACE_ERROR_RETURN ((LM_ERROR, "%p\n", "open"), -1); /* Create a simple buffer to receive the data. You generally need to provide a buffer big enough for the largest datagram you expect to receive. Some platforms will let you read a little and then some more later but other platforms will throw out whatever part of the datagram you don't get with the first read. (This is on a per-datagram basis BTW.) The theoretical limit on a datagram is about 64k. The realistic limit (because of routers & such) is much smaller. Choose your buffer size based on your application's needs. */ char buf[BUFSIZ]; /* Unlike ACE_SOCK_Stream, datagrams are unconnected. That is, there is no "virtual circuit" between server and client. Because of this, the server has to provide a placeholder for the OS to fill in the source (client) address information on the recv. You can initialize this INET_Addr to anything, it will be overwritten when the data arrives. */ ACE_INET_Addr remote; ACE_DEBUG ((LM_DEBUG, "(%P|%t) starting up server daemon\n")); /* Receive datagrams as long as we're able. */ while (dgram.recv (buf, sizeof (buf), remote) != -1) { /* Display a brief message about our progress. Notice how we use the 'remote' object to display the address of the client. With an ACE_SOCK_Stream we used get_remote_addr() to get the address the socket is connected to. Because datagrams are unconnected, we use the addr object provided to recv(). */ ACE_DEBUG ((LM_DEBUG, "(%P|%t) Data (%s) from client (%s)\n", buf, remote.get_host_name ())); /* To respond to the client's query, we have to become a client ourselves. To do so, we need an anonymous local address from which we'll send the response and a datagram in which to send it. (An anonymous address is simply one where we let the OS choose a port for us. We really don't care what it is. */ ACE_INET_Addr local ((u_short) 0); ACE_SOCK_Dgram client; /* Open up our response datagram as always. */ if (client.open (local) == -1) { ACE_ERROR_RETURN ((LM_ERROR, "%p\n", "client open"), -1); return 0; } /* Build a witty response... */ sprintf (buf, "I am here"); /* and send it to the client. Notice the symmetry with the recv() method. Again, the unconnected nature of datagrams forces us to specify an address object with each read/write operation. In the case of read (recv()) that's where the OS stuffs the address of the datagram sender. In the case of write (send()) that we're doing here, the address is where we want the network to deliver the data. Of course, we're assuming that the client will be listening for our reply... */ if (client.send (buf, ACE_OS::strlen (buf) + 1, remote) == -1) ACE_ERROR_RETURN ((LM_ERROR, "%p\n", "send"), -1); } return 0; }
And that's really all there is to it. Obviously there is some room for improvement. The most blatant is the somewhat small buffer size for receiving the datagram. I've never been able to get a solid answer on datagram sizes. The theoretical limit is just under 64k but you have to deal with fragmentation. Some readings indicate that 8k is a reasonable size, others go much smaller. My general rule of thumb is to keep datagrams relatively small (eg -- under 8k or so) and test a lot. If you find that your routers are fragmenting your larger datagrams, back off to something smaller. Of course, if you must send 100k and can only do so 1k at a time, you'll have to worry about retransmissions & reordering. At that point, you might consider going to TCP. Remember: datagrams are unreliable! Don't try to make 'em do something they werent' designed for!