routing table wierdness

[Hans Breitenlohner]

In article <1990Jul31.184659.6260 at wrl.dec.com> mogul at wrl.dec.com 
(Jeffrey Mogul) writes:
=
=I didn't examine the routing tables in the original message that
=carefully, so they could have reflected a problem of some sort ...
=but please note that there is nothing intrinsically wrong with
=having your host's routing tables built up out of redirects.
=
=In fact, this is precisely the RIGHT way for things to work, under
=current thinking.  Hosts should not know about routing protocols
=or routes to non-local networks; rather, a host should know the
=address of one or more "default routers", and for non-local destinations
=should pick one of its default routers more or less at random (there
=is a concept of "preference level" to confuse the issue here, but
=one can ignore that in most cases).  The routers know the right routes,
=and if a host makes the wrong choice, the router in question provides
=an ICMP Redirect referring to the specific destination host.
=
...
=
=-Jeff

There is a problem with this approach in connection with the way
Ultrix currently works:  Once you have a redirect, it will be there
forever (unless you manually flush the routing tables, or you run gated).
If the entry should be bad, either because it was generated by a transient
routing disturbance, or because the network configuration has changed,
then that host may be permanently unreachable.  Since other hosts
on the same network will work fine, such problems are great fun to debug.

Hans