TcpWin tunning:
===============

TCP windowing the part of the TCP protocol that defines about how
much bytes could measure each packet sent or recived using a socket
connection.

This parameter is fetched using getsockopt and getting the RCVBUF
argument. And SNDBUF for output buffer.

A simple hack on that could be to force this values, and use the
transfer media faster. The base problem comes when we give a bigger
value to the tcp window than the allowed by the fisic media. Then
each packet must be re-send, causing an ugly  payload.

Under Linux this value could be changed using:

# sysctl -w net.ipv4.route.max_size=??


Under NetBSD you can use:

# sysctl -w net.inet.tcp.sendspace = ?
# sysctl -w net.inet.tcp.recvspace = ?
# sysctl -w net.inet6.tcp6.sendspace = ?
# sysctl -w net.inet6.tcp6.recvspace = ? 

Also you have the *.win_scale, *init_win that could be also hacked.

Another way to do this, is coding inside sws a simple setsockopt
that forces a new sendbuf/recvbuf on the desired socket.

This could be also used as a simple QoS that could be nice to 
implement in a short future, using (rcv/snd)buf and usleeps time
between each packet.
