jnford@jay.weeg.uiowa.edu (Jay Ford) (07/26/90)
We currently have a class-B number (128.255) with 8-bit subnetting (mask 255.255.255.0). Some folks on our campus net have an interest in subdividing a subnet into a bunch of 11-bit sub-subnets (mask 255.255.255.224). The sub-subnets will each contain a small number of systems (< 32), so full 8-bit subnets would be sparsely used. This probably requires a router which has interfaces with different masks, one with 255.255.255.0 and one with 255.255.255.224. We'd like to use a BSD system as a router. I don't think routed (the BSD RIP implementation) will deal with the inconsistent masks, and it doesn't seem that gated likes them very well either. I'd appreciate some advice from anyone who's done this sort of thing. It seems like a fairly reasonable way to better utilize the address space, but the standard software doesn't appear to support it. Thanks. Jay Ford, Weeg Computing Center, University of Iowa, Iowa City, IA 52242 jnford@jay.weeg.uiowa.edu or jnfordpb@uiamvs.bitnet, 319-335-5555
kwe@bu-it.bu.edu (Kent England) (07/26/90)
In article <1951@ns-mx.uiowa.edu>, jnford@jay.weeg.uiowa.edu (Jay Ford) writes: > We currently have a class-B number (128.255) with 8-bit subnetting (mask > 255.255.255.0). Some folks on our campus net have an interest in subdividing a > subnet into a bunch of 11-bit sub-subnets (mask 255.255.255.224). The > sub-subnets will each contain a small number of systems (< 32), so full 8-bit > subnets would be sparsely used. > > This probably requires a router which has interfaces with different masks, one > with 255.255.255.0 and one with 255.255.255.224. We'd like to use a BSD system > as a router. I don't think routed (the BSD RIP implementation) will deal with > the inconsistent masks, and it doesn't seem that gated likes them very well > either. > > I'd appreciate some advice from anyone who's done this sort of thing. It seems > like a fairly reasonable way to better utilize the address space, but the > standard software doesn't appear to support it. > Most routing protocols don't carry the subnet mask along with the routes in the update messages. They assume that there is one consistent subnet mask throughout the entire subnetted net. RIP is like that. OSPF does understand variable length subnet masks and OSPF is in the process of being implemented in parts of the Internet, so you have some help and guidance from Milo Medin at NASA Ames and Dave O'Leary of SURAnet to name two I know. I've designed a variable length subnet mask approach for OSPF, but I haven't implemented it. I think it can be managed in a relatively straightforward way, if you can explain the proper configurations to your user community. Otherwise, rely on trusty proxy ARP. :-) The gated people are working on OSPF for gated, so if you can wait a little you can have a protocol that will run on a BSD system. If your friends were trying to make bigger, not smaller, subnets then you could use the trick of layering several 8-bit subnets onto one interface. cisco calls their configuration "secondary interface", for example. It's wise to tell all hosts on layered subnets that their mask is 255.255.0.0 and rely on proxy ARP. Otherwise, suffer an extra hop thru the local router. I doubt you would want to change your default subnet mask to 11 bits and do a lot of layering to get 8 bit subnets. This sort of trick is nice if you don't want to go to OSPF or if you are migrating from a bridged to mixed environment. AppleTalk DDP-IP gateways like the FastPath and the GatorBox can use transparent subnetting, where a group of addresses are assigned to the gateway for the LocalTalk network. I don't know a way to exploit that for you. --Kent