HP-UX TCP/IP Performance White Paper, March 2008

sequence number space within 2MSL. The suggested-by RFC's value for 2MSL would be four minutes, hence

timestamps should be used whenever a TCP connection is going to run at sustained rates of more than 1 GB

per minute, which is ~17 MB/s or ~145 Mbit/s. Admittedly, this is pretty conservative.

tcp_tw_cleanup_interval:

Interval in milliseconds between checks to see if TCP connections in

TIME_WAIT have reached or exceeded the tcp_time_wait_interval.

Reducing the length of this interval increases the precision of the

timeout.

If tcp_time_wait_interval ia set to 60 seconds (the default) and the

tcp_tw_cleanup_interval is also 60 seconds, a TCP connection may spend

as little as 60 seconds, or as much as two minutes in TIME_WAIT.

[10000-300000] Default: 60000

tcp_xmit_hiwater_def:

The amount of unsent data that triggers write-side flow control.

[4096,-] Default: 32768 bytes

This can be thought of as the default send socket buffer size (SO_SNDBUF). Once this many bytes of data

are queued to the connection (actually, it tends to be more than this value) any subsequent send() (etc)

call with either block or return EWOULDBLOCK/EAGAIN if the socket was marked non-blocking.

Increasing this value allows an application to put more data into the connection at one time. This can be

useful to minimize the number of send()/write() system calls required to get the data into the transport.

Also, it allows the transport a better chance at taking full advantage of the remote TCP's advertised

window, which might otherwise be limited by flow control on the sending application. See the discussion of

tcp_recv_hiwater_def for a discussion of the limit to throughput based on window size and RTT - the

same idea holds.

Having this greater quantity of data in flight allows greater throughput in the face of occasional packet loss.

There is a slight chance that too large a value will put more data out onto the network than the network can

hold and actually cause packet losses.

tcp_xmit_hiwater_lfp:

The amount of unsent data that triggers write-side flow control

for fast links.[4096,-] Default: 65536 bytes

tcp_xmit_hiwater_lnp:

The amount of unsent data that triggers write-side flow control

for slow links.[4096,-] Default: 8192 bytes

tcp_xmit_hiwater_max:

Limits the send buffer size for TCP sockets or communication

endpoints specified in a SO_SNDBUF option of a setsockopt() call

or XTI_SNDBUF option in a t_optmgmt() call.