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

tcp_ip_ttl:

TTL value inserted into IP header for TCP packets only.

[1, 255] Default: 64

A default value of 64 means that TCP will not communicate with any system that is more than 64 hops

(routers) away. This should be sufficient for 99.999% of all cases. However, increasing this value to 255

would have no downside in a non-error case. It would make things a bit worse in the case of a routing loop

(router A sends to router B which sends back to router A and the like) as it would take that many more trips

through the loop to "kill-off" the IP datagram containing that TCP segment.

tcp_ip6_hop_limit:

Hop-Limit value inserted into IPv6 header for TCP packets only.

[1, 255] Default: 64

tcp_keepalive_detached_interval:

Interval for sending keep-alive probes when TCP is detached.

When a TCP instance exists without a corresponding stream or

queue, that TCP instance is referred as "detached". When a stream

closes before the TCP FIN-ACK handshake is complete, TCP must

retain information stored in its instance data to complete the

handshake.

If the ACK to terminate a connection in FIN_WAIT_2 state never

arrives, then a detached TCP instance could remain allocated

indefinitely with no way to delete it. To avoid this potential

problem, a keep-alive timer is started when a TCP stream closes

before the TCP state is CLOSED.

The time-out interval is tcp_keepalive_detached_interval. If this

timer expires, the keep-alive algorithm will ensure that the TCP

instance will eventually be deleted.

[10000,10*24*3600000] Default: 120000 (2 minutes)

Here "without a stream or queue" (ie detached) in essence means "no socket." That is to say, the

application has close()'d the socket.

This value determines when keepalive probes will start for a detached connection in FIN_WAIT_2. So long

as the remote (or something taking responsibility for the remote - for instance a firewall) responds to the

keepalive probes, the connection will remain – after all, if the firewall is still responding when the remote is

really gone, the firewall is broken and must be fixed.

Once keepalive probes start, they will be retransmitted, and the connection possibly aborted, based on the

regular TCP retransmission timer settings.

tcp_keepalive_detached_interval is only for detached connections in the FIN_WAIT_2 or

LAST_ACK state. The classic TCP keepalive mechanism for connections in other active states and enabled

by a call to setsockopt() is controlled via tcp_keepalive_interval.