Should routers send redirects by default?

Wed Aug 25 06:03:31 UTC 2010

> > Forgetting all of the theoretical constructs for a moment, has anyone
> > here personally encountered an operational scenario in which ICMP
> > redirects solved a problem for you that you would otherwise have found
> > difficult or intransigent? Without naming names, would you describe
> > the scenario's details, explain the problem that would have existed
> > absent redirects and explain how redirects solved it for you?
> 
> I've never had redirects solve a problem for me.

Once upon a time, gatekeeper.dec.com was a MIPS Ultrix box connected
to a LAN segment (the IP prefix for that LAN segment was 16.1.0.0/24,
and gatekeeper used to be 16.1.0.2). Also connected to the LAN segment
were routers belonging to AlterNet (or maybe it was UUnet by then) and
BARRnet.

gatekeeper's static default route was pointed at the BARRnet
router. The BARRnet and AlterNet routers exchanged routes out of sight
of the LAN segment in question. When BARRnet knew that the destination
of a packet would be the AlterNet router, it would issue an ICMP
redirect to gatekeeper (or any of the other hosts that pointed default
at the BARRnet router). The redirects let us make pretty efficient use
of the interfaces toward both from the collection of gatekeeper and
uucp-gw{1,2} and the NNTP relays and such. Had we shovelled all the
traffic at BARRnet, all the time, we wouldn't have stretched our
DELNIs as far as we did.

That was 1994 - and in fairly short order (within two years) we went
from DELNIs to DECswitch 900s, MIPS Ultrix to Digital UNIX (or
whatever it was called) on AlphaStations (so many AlphaStations), and
ICMP redirects to the private IX that became PAIX. 

Stephen