On 2013-01-09 19:20, Peter Lothberg wrote: Also, the metrics are set on the circuit, and not individual destinations. While this worked more or less ok for DEC back in the day, with the bridge, the cost of two different destinations over the same circuit could in reality be very different. Bridged ethernet is a BAD thing for most protocols. And if you use spanning tree, it's not making optimal use of the topology. Totally agree. Radia finally came around and "Trill" is a routed network to carry ethernet frames. Hmm. Interesting. Might be something to look at, if I ever get plenty of time... As the cost/hop is in the routing vectors being sent periodiacally, you could implement a "cost/hop" NAT (half nat, one half on send one half on receive and some protocol to exchange how you want the metric matrix to look). That could be interesting, but since I don't expect us to fool around with the DECnet protocols as such, I guess this will not happen. Using dedicated routers that can tunnel traffic is really the best solution. Be it Ciscos, Multinet on VMS, or Rob's user level router... I think we're slowly getting more in that direction. Johnny