> On Wednesday 02 February 2005 16:57, Hervé wrote: > >> Basically, the network interface where the packet originated has to >> somehow be remembered at the routing stage. I only see source address >> mangling to do that... >> > > It would seem that if you add two additional IPs to the NFS machine (on a > dummy interface) and do a local DNAT to those IPs based on the incoming > nic, > that the correct routing tables/rules would be all that is needed to route > the return packets to the correct network. > > Have a look at > > http://www.policyrouting.org/PolicyRoutingBook/ONLINE/TOC.html > > for the concepts and examples. Chapter 5, in particular, covers the > routing > concepts your solution will need. > > Note that ProxyARP uses the same ip on two interfaces and depends on the > proper routing table entries to do the right thing. > > > -- > > Bob Tellefson > Java network application development & hosting > Yup, that's what I started doing, but I still have to modify the source address somehow, so the return packet is routed according to that address... In fact, conceptually, I really want a router between each client and my server: Client1 (a.b.c.d) -> (a.b.c.1) Router (x.y.z.1) -|-> (x.y.z.3) Server Client2 (a.b.c.d) -> (a.b.c.1) Router (x.y.z.2) -| This does all: same client IPs, unique server for all the clients IP (and only one NIC). And that's probably what I'm going to do, not just conceptually! But I have to say that I would have liked to do it all on the one machine, rather than needing to go through routers. Thanks for the link as well, I need to read through it carefully before giving up my first idea (I hate giving up ;-) Herve.
