HP-UX 11i June 2002 Release Notes

Internet and Networking Services
Network Transport
Chapter 12
215
Specific Changes
1. All 1’s or all 0’s are now allowed in masked bits of subnet address:
ip_check_subnet_addr shows whether or not that RFC1122 or RFC1878 enforces
the network subnet mask. If it is a 0 (zero), then the RFC1122 behavior is seen. If it
is a 1 (one), then RFC1812 is seen. The default is now RFC1812 behavior. (See
ifconfig in the preceding section for more information.)
This new behavior, an enhancement, makes available up to twice as many IP
addresses than a similarly configured RFC1122 machine.
2. TCP-supported tcp_sack_enable now enables selective acknowledgement. This
enhancement could improve performance in networks with large transmission
windows by allowing TCP recipients to indicate lost segments within large
transmission blocks. The TCP sender can then retransmit only the lost segments.
Supported parameter values are as follows:
2 Local system enables SACK if remote system first sends SACK
(Default).
1 Local system requests the SACK option during a connect()
request.
0 Local system never uses SACK.
3. Send and receive buffers are now limited by the following tunable:
Specifically, udp_recv_hiwater_max (default 2 GB) sets the maximum receive buffer
size that setsockopt or t_optmgmt can set for a UDP socket. The TCP version of
this is tcp_recv_hiwater_max (default 1 GB). On the other hand,
tcp_xmit_hiwater_max (default 2 GB) sets the maximum send buffer size that
setsockopt or t_optmgmt can set for a TCP socket. There is no UDP version of this
value. These system-wide parameters prevent processes from keeping large amounts
of data in send or receive buffers, and thereby consuming system resources.
CAUTION Although the enhancement tcp_fin_wait_2 may be used to set how long a
connection will be in FIN_WAIT_2, you should use this cautiously. If the remote TCP
entity is slow, but would terminate normally (is not hung nor will terminate
abnormally), TCP may close the connection prematurely. This could result in the
flushing of data in the remote connections receive buffer. If this happens
unexpectedly, then the data could become corrupted.
For more information on using tcp_fin_wait_2, enter the following command:
ndd -h tcp_fin_wait_2.
With a new field, ip_udp_status now reports how many times a given UDP socket
has overflowed. Although this enhancement only works on sockets currently open, it
can be a very handy troubleshooting tool used when netstat -p udp shows socket
overflows.
4. With a TimeStamps option now supported, tcp_ts_enable allows RFC 1323
TimeStamp extensions to TCP Headers. The TimeStamps are used for two purposes:
a. RTTM (Round Trip Time Measurement) of the interval between the time a TCP
sends a segment and the time the return acknowledgement arrives.
b. PAWS (Protect Against Wrapped Sequences) on high-speed networks.