ttcp -- Transmission Control Protocol Extensions for Transactions
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
int
setsockopt(sock, IPPROTO_TCP, TCP_NOPUSH, &One, sizeof One);
ssize_t
sendto(sock, msg, len, MSG_EOF, &sin, sizeof sin);
ssize_t
sendto(sock, msg, len, MSG_EOF, 0, 0);
T/TCP refers to a set of extensions to the TCP protocol (see tcp(4))
which permit hosts to reliably exchange a small amount of data in a twopacket
exchange, thus eliminating the extra round-trip delays inherent in
a standard TCP connection. The socket interface includes modifications
to support T/TCP, detailed here for the specific case, and in the
socket(2) and send(2) manual pages for the protocol-independent support.
T/TCP is defined in RFC 1644.
The T/TCP extensions work by including certain options in all segments of
a particular connection, which enable the implementation to avoid the
three-way handshake for all but the first connection between a pair of
hosts. These same options also make it possible to more reliably recognize
old, duplicate packets, which in turn reduces the amount of time the
TCP protocol must maintain state after a connection closes. The
``net.inet.tcp.rfc1644'' MIB variable can be used to disable T/TCP negotiation
at run time; however, the protocol has been designed to ensure
that attempts by non-T/TCP systems to communicate with T/TCP-enhanced
ones automatically degenerate into standard TCP.
The expected model of a ``transaction'' as used by T/TCP is a fairly simple
one:
1. A client program generates a request to be sent to the server, which
is small enough to fit in a single TCP segment, and sends a SYN PUSH
FIN segment with options and data to the server.
2. The server program accepts the request in the same manner as for
regular TCP connections, interprets it, and generates a reply which
may be small enough to fit in a single segment. If it is, the reply
is sent in a single SYN PUSH FIN ACK segment with (different)
options and data back to the client. If not, then the connection
degenerates into (almost) the usual case for TCP. The server then
closes its socket.
3. The client reads the reply and closes its socket.
Support on the client side is provided by extending the semantics of the
sendto(2) and sendmsg(2) system calls to understand the notion of
``implied connect'' and ``send and shutdown''. To send the request in a
transaction, the sendto(2) system call is typically used, as in the following
example:
char request[REQ_LEN];
struct sockaddr_in sin;
int sock, req_len;
sock = socket(PF_INET, SOCK_STREAM, 0);
/* prepare request[] and sin */
err = sendto(sock, request, req_len, MSG_EOF,
(struct sockaddr *)&sin, sin.sin_len);
/* do something if error */
req_len = read(sock, request, sizeof request);
close(sock);
/* do something with the reply */
Note that, after the call to sendto(), the socket is now in the same
state as if the connect(2) and shutdown(2) system calls had been used.
That is to say, the only reasonable operations to perform on this socket
are read(2) and close(2). (As the client's TCP sender is already shut
down, it is not possible to connect(2) this socket to another destination.)
There are two different options available for servers using T/TCP:
1. Set the TCP_NOPUSH socket option, and use normal write(2) calls when
formulating the response.
2. Use sendto(2) with the MSG_EOF flag, as in the client, but with the
destination unspecified.
The first option is generally the appropriate choice when converting
existing servers to use T/TCP extensions; simply add a call to
setsockopt(sock, IPPROTO_TCP, TCP_NOPUSH, &One, sizeof One) (where One is
an integer variable with a non-zero value). The server socket must be
closed before any data is sent (unless the socket buffers fill up).
The second option is preferable for new servers, and is sometimes easy
enough to retrofit into older servers. In this case, where the reply
phase would ordinarily have included a call to write(), one substitutes:
sendto(sock, buf, len, MSG_EOF, (struct sockaddr *)0, 0)
In this case, the reply is sent immediately, but as in the client case,
the socket is no longer useful for anything and should be immediately
closed.
The T/TCP extensions require the ``net.inet.tcp.rfc1644'' MIB variable to
be true in order for the appropriate TCP options to be sent. See tcp(4)
for more information.
send(2), setsockopt(2), inet(4), tcp(4)
R. Braden, T/TCP - TCP Extensions for Transactions, RFC 1644.
Support for T/TCP first appeared in FreeBSD 2.1, based on code written by
Bob Braden and Liming Wei at the University of Southern California,
Information Sciences Institute, and ported by Andras Olah at the University
of Twente.
FreeBSD 5.2.1 January 18, 1995 FreeBSD 5.2.1 [ Back ] |
