sater@cs.vu.nl (Hans van Staveren) (10/20/89)
I just noticed that on our Suns, running SunOS 4.0.3 dynamically added routes are kept infinitely. This runs you into problems when gateways change. Is this up-to-date software behaviour? A little thought suggested the following protocol to me: Each dynamically added route(by an ICMP redirect) is aged after a reasonable time(10 minutes?) and the first packet after the age is sent to the default router. Any new ICMP redirect for that route replaces the current one, if no new one comes, when the gateway is down for example you keep using the old one and try the default route every say 100 packets. Does this seem reasonable? Is something like this already in newer software? Hans van Staveren, Vrije Universiteit, Amsterdam, Holland
smb@ulysses.homer.nj.att.com (Steven M. Bellovin) (10/21/89)
The host requirements RFCs addresses this point, but the issue is far from settled. Basically, they call for hosts to implement some sort of dead gateway detection code; the route should be squashed at such times. If the gateways, negotiating amidst themselves, decide that a different gateway is now better -- and each gateway is assumed to know such things -- then the old gateway will send out a new redirect message. (Dead gateways can be detected by things like the sudden absence of ACKs to TCP.) Personally, I'm not completely convinced. Dead gateway detection is often hard, and other factors, notably security considerations, make me less than fond of redirects in the first place. But the RFCs are certainly the place to start.