TCP Retransmit counts

[Peter Jeremy]

I often need to move my laptop around the house and have temporary
network outages (no wireless).  I've noticed that any active TCP
connections drop out fairly quickly - some checking with tcpdump shows
that there is only 60 seconds between the first transmit attempt and
the last re-transmit.  This strikes me as excessively short - it's
likely to take more time than this for a managed switch or router
to reboot.

I've previously (5-6 years ago) done some testing on other commercial
Unices and got figures of 8-10 minutes - which seems more reasonable.

Having had a look at the code, there are TCP_MAXRXTSHIFT (12)
re-transmit attempts with exponential back-off based on the calculated
RTT.  Whilst this is probably reasonable for a WAN link, I would like
to extend the timeout on LAN links.

Can anyone see any downsides to either increasing TCP_MAXRXTSHIFT (and
associated timer arrays) or changing retransmit timeout to having a
minimum value (similar but opposite to the tcp_maxpersistidle test in
tcp_timer_persist)?

-- 
Peter Jeremy
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-net/attachments/20060802/66fb9259/attachment.pgp