IPv6 Routing Header - How does it work?

[mcmanus at ducksong.com (Patrick R. McManus)]

[Vriz: Thu, Apr 18, 2002 at 09:26:55AM +0800]
> Thanks Patrick. May I know if there's any way (using ICMP?) for me at the
> source node to find out what is the shortest route to the destination node?
> I mean the intermediate nodes addresses.
> Thank you.

The basic answer to your question is no. First of all, "shortest" can
be defined many different ways (fiber miles, miles weighted by medium,
congestion, link latency, link bandwidth, etc..) More importantly,
policy plays a very strong role in route decisions as well (cheapest
link, links that need balanced traffic ratios, hauling to peer vs
hot-potato transit, etc..) 

And each network that your packet travels makes all of the above
decisions independently of the other networks.. This makes predicting
paths (much less optimal paths!) very hard. Indeed, the nature of
stateless routing says that when one router passes the packet on he
has no idea what the next one will do with it - there is no concept of
end-to-end path from a router's point of view - only next hop.

you might make some progress by searching for "bgp looking glass" on
google.

you seem to be implying that you want to use loose source routing to
manually select a more efficient path for your traffic. Don't do that
- it won't work for numerous reasons. Even if you weren't filtered
(any you likely would be - by either a customer border router or the
host itself), static routes on a global scale are a very bad
idea. They can't react to changes in topology caused by severe
congestion or reachability - that's why we have dynamic routing
protocols like bgp, is-is and ospf. source routes only really make
sense as a diagnostic tool.

at this point though, this doesn't have much to do with linux or unix.

-Patrick