cudcv@warwick.ac.uk (Rob McMahon) (05/14/91)
We have a class B address (137.205.0.0) for the University. Currently the whole campus is running without subnets, with bridges between each department and the backbone, but we allocated the numbers such that we could later give each department its own subnet, and change the bridges for routers should it become necessary. We now have our first department that actually wants to be split off with a router, and I realise that I don't understand how it works at all. Here was the plan (internet address/netmask): 137.205.0.0/0xffff0000 | | ---------------------------- | | Bridge Router | | 137.205.232.0/0xffff0000 137.205.176.0/0xfffffc00 But the Cisco router we had on trial won't let you set different netmasks on its two interfaces "in the current implementation". Does anyone know if this is likely to change ? A nearby book calls what we are trying to do an "illegal" setup, and says it is "recommended" to have the same netmask throughout the network. It seems that to make this work the routing tables on the Unix hosts ought to have a netmask associated with each entry, but it's not there. When I tried to add a route to such a gateway the machine lost contact with the entire world. (I still don't understand this, if the machine was just picking the wrong route and sending all packets to the router, why didn't the router just forward them to the correct machine ?) Is this setup not supposed to work ? Why not ? It seems like an obvious application (so obvious that I just assumed it was going to work without really sitting down and thinking about it). We don't want to have to subnet the entire University just yet, but it would be nice to keep our options open. Cheers, Rob -- UUCP: ...!mcsun!ukc!warwick!cudcv PHONE: +44 203 523037 JANET: cudcv@uk.ac.warwick INET: cudcv@warwick.ac.uk Rob McMahon, Computing Services, Warwick University, Coventry CV4 7AL, England
lars@spectrum.CMC.COM (Lars Poulsen) (05/17/91)
In article <=~P_0Q&@warwick.ac.uk> cudcv@warwick.ac.uk (Rob McMahon) writes: >We have a class B address (137.205.0.0) for the University. Currently the >whole campus is running without subnets, with bridges between each department >and the backbone, but we allocated the numbers such that we could later give >each department its own subnet, and change the bridges for routers should it >become necessary. > >... our first department ... actually wants to be split off with a router, >... Here was the plan (internet address/netmask): > > 137.205.0.0/0xffff0000 > | > | > ---------------------------- > | | > Bridge Router > | | > 137.205.232.0/0xffff0000 137.205.176.0/0xfffffc00 > >It seems that to make this work the routing tables on the Unix hosts ought to >have a netmask associated with each entry, but it's not there. Briefly: Yes, it will not work, for the reason stated. The next generation of IP routers will keep masks with routes at all times, but the routing protocol software to distribute and manage such routes is still not frozen solid enough to give to the unwashed masses (OSPF-2 is in draft, and there is still a vocal minority that insists that "dual IS-IS" MUST be supported). Universally masked routes will also resolve the other thing that "intuitively ought to work": Disjointed subnets. -- / Lars Poulsen, SMTS Software Engineer CMC Rockwell lars@CMC.COM
kre@cs.mu.oz.au (Robert Elz) (05/20/91)
cudcv@warwick.ac.uk (Rob McMahon) writes: >A nearby book calls what we are trying to do an "illegal" setup, and says it >is "recommended" to have the same netmask throughout the network. Its not illegal, but its not supported by just about anything yet. However, support for this isn't as hard as was implied by a previous response - routers do need to keep the network mask with every route, but its not essential to use a routing protocol that transmits masks, RIP can cope in this kind of evnironment just fine (and if RIP does, so will just about anything else reasonable - just leave out EGP). What a new routing protocol is needed for is the ability to assign the subet masks in any random fashion - but is you're willing to assign netmasks, and net numbers, in a structured fashion with respect to the physical topology of the net, then it will all work just fine. That restriction is typically not a problem at all in most environments, esp where you have a backbone with a wide subnet mask, and spur nets with narrow masks. If only you could find routers to support it - its not difficult, masks in the routing table, and a little proxy ARP is all that's required. kre