Measure UDP send and receive performance.  Be very careful that the
data is reliable.  Avoid unnecessary system calls that may affect
the result.

The command line tool udpbench can operate either in send or receive
mode.  It operates time based with an alarm signal.  In send mode
it calls send(2) until the alarm triggers.  When receiving, the
recv(2) loop terminates after the timeout, which should be larger
than the send timeout.  On top there is an idle detection to terminate
receive mode prematurely.

To calculate the send throughput, gettimeofday is called before the
first and after last send(2) system call.  The number of successful
system calls is the assumed the number of packets.

On the receive side gettimeofday is called after the first recv(2)
system call returns.  All other recv(2) calls are blocking for at
most 100 milli seconds.  If there was no packet received for one
second or if the timeout fires, the recv(2) loop terminates.  The
later case is considered as an error.  The thoughput is calculated
by packets successfully received and by the time between the return
of the first and the last successful recv(2).

To coordinate both sides, a locally running udpbench can automatically
start a udpbench on a remote machine via ssh.  The remote send or
receive mode is reversed.  All size, length or timeout parameters
are used for both sides.  The hostname or IP address must be specified
for the receiving machine.