TCP/UDP benchmarking and measurement
tcpbench is a small tool that performs
throughput benchmarking and concurrent sampling of kernel network
tcpbench is run as a client/server pair.
The server must be invoked with the
-s flag, which
will cause it to listen for incoming connections. The client must be invoked
with the hostname of a listening server to connect
Once connected, the client will send TCP or UDP traffic as fast as
possible to the server. Both the client and server will periodically compute
and display throughput statistics. The server starts computing these for UDP
on receipt of the first datagram, and stops for TCP when it has no
connections. This display also includes any kernel variables the user has
selected to sample (using the
-k option, which is
only available in TCP mode). A list of available kernel variables may be
obtained using the
A summary over the periodic throughput statistics is displayed on
exit. Its accuracy may be increased by decreasing the
interval. The summary bytes and duration cover the
interval from transfer start to process exit. The summary information can
also be displayed while
tcpbench is running by
sending it a
SIGINFO signal (see the
status argument of
stty(1) for more information).
The options are as follows:
tcpbenchto use IPv4 addresses only.
tcpbenchto use IPv6 addresses only.
- Specify the size of the internal read/write buffer used by
tcpbench. The default is 262144 bytes for TCP client/server and UDP server. In UDP client mode this may be used to specify the packet size on the test stream.
- Specify the IP address to send the packets from, which is useful on machines with multiple interfaces.
- Enable debugging on the socket.
- Specify one or more kernel variables to monitor; multiple variables must be separated with commas. This option is only valid in TCP mode. The default is not to monitor any variables. Using this option requires read access to /dev/kmem, which may be restricted based upon the value of the kern.allowkmem sysctl(8).
- List the name of kernel variables available for monitoring and exit.
- Use the given number of TCP connections (default: 1). UDP is connectionless so this option isn't valid.
- Specify the port used for the test stream (default: 12345).
- In client mode the write buffer size is randomized up to the size
- Specify the statistics interval reporting rate in milliseconds (default: 1000). If set to 0, nothing is printed.
- Set the size of the socket buffer used for the test stream. On the client this option will resize the send buffer; on the server it will resize the receive buffer.
tcpbenchin server mode, where it will listen on all interfaces for incoming connections. It defaults to using TCP if
-uis not specified.
- Change the IPv4 TOS or IPv6 TCLASS value. toskeyword may be one of critical, inetcontrol, lowdelay, netcontrol, throughput, reliability, or one of the DiffServ Code Points: ef, af11 ... af43, cs0 ... cs7; or a number in either hex or decimal.
- Stop after secs seconds.
- Use AF_UNIX sockets instead of IPv4 or IPv6 sockets. In client and server mode hostname is used as the path to the AF_UNIX socket.
- Use UDP instead of TCP; this must be specified on both the client and the
server. Transmitted packets per second (TX PPS) will be accounted on the
client side, while received packets per second (RX PPS) will be accounted
on the server side. UDP has no Protocol Control Block (PCB) so the
-kflags don't apply.
- Set the routing table to be used.
- Display verbose output. If specified more than once, increase the detail of information displayed.
tcpbench program first appeared in
tcpbench program was written by
UDP mode and libevent port by Christiano F. Haesbaert <email@example.com>.