garethh@sadss (Gareth Howell ) (06/12/91)
Hi all, I wonder if someone can help me with a tricky problem I am trying to come to grips with. I have the (un)enviable taks of coming up with an Internet Addressing Strategy for the UK Department of Social Security's internet (note small 'i':-). This comprises (or soon will) 2500+ Ethernet LANs: each with anything from 4-50 PCs, Unix application servers and gateways on them; all interconnected using the Government Data Network (X.25 (1980)). Most of the operational systems use OSI protocols, but there is a significant amount of IP traffic, mainly for SNMP HUB and BRIDGE and host management on the LANs. What I need is a sanitory way to split up the population to ease number allocation and permit local administration of each LAN. What I have come up with is this, and I would like comments (good or bad :-): Allocate a single non-subnetted Class B address to the X.25 GDN (2500+ hosts). Allocate a number of Class B addresses to clusters of LANs, and subnet each of these networks in accordance with RFC950 and RFC1219. I have one outstanding issue relating to this, and that is whether dynamic routing protocols will cope with this environment. Specifically, will a host on LAN 'A' (which is a subnet of network 'X') be able to reach a host on LAN 'B' (which is a subnet of network 'Y') by routing across the GDN. The problem seems to be whether the routing tables in LAN 'A's GDN gateway, know that to get to LAN 'B' you have to go to LAN 'B's GDN gateway: which implies that LAN 'A's gateway (on network 'X'), knows the subnet mask of network 'Y'. I'm not sure this is possible; but the alternatives of allocating a single Class A network address to cover the lot, or allocate 2500+ Class C addresses + 1 Class B address (for the GDN) are either impractical, or unmanageable (and anti-social if we advertise 2500+ networks to the core!!! ). Any ideas? Gareth Howell garethh%sadss.uucp@ukc.ac.uk (I think that's the best route for mail)
lars@spectrum.CMC.COM (Lars Poulsen) (06/14/91)
In article <m1TF42w163w@sadss> garethh@sadss (Gareth Howell ) writes: >I have the (un)enviable taks of coming up with an Internet Addressing >Strategy for the UK Department of Social Security's internet (note small >'i':-). This comprises (or soon will) 2500+ Ethernet LANs: each with >anything from 4-50 PCs, Unix application servers and gateways on them; all >interconnected using the Government Data Network (X.25 (1980)). Most of the >operational systems use OSI protocols, but there is a significant amount of >IP traffic, mainly for SNMP HUB and BRIDGE and host management on the LANs. > >What I need is a sanitory way to split up the population to ease number >allocation and permit local administration of each LAN. What I have come up >with is this, and I would like comments (good or bad :-): > >Allocate a single non-subnetted Class B address to the X.25 GDN (2500+ hosts). >Allocate a number of Class B addresses to clusters of LANs, and subnet each >of these networks in accordance with RFC950 and RFC1219. I am involved with defining a similar network in this country. Ours is worse, in that most of the X.25 connections are dial-up on-demand (inbound-only). The proposed solution in our network is to hierachically define all of the address space in a number of class B networks, one per region, with the physical X.25 WAN appearing in segments of each class B net. The regions are each headed up by an IP router, and these routers (which will probably all be co-located with a major hub in the X.25 network) will be connected via a backbone LAN. Each leaf PC only needs a default route to the Ether side of the leaf gateway. Each leaf gateway only needs a default route via X.25 to the regional router. The regional router needs to know the subnet numbers, subnet masks and X.25 addresses of all the ethernet segments within the region. But anything outside the region can be routed by net number (without subnets). To make this work, an IGP must be used which can communicate masks with all routes, but the whole cluster can connect to the outside world by EGP and only tell the core about the regional class B numbers. Note, that things would have been defined very differently if the wide area network had been managed primarily for IP use and had had an established DDN-like IP-to-X.25 address mapping. I hope this is helpful. Feel free to contact me by email. The above is probably as much as is useful to the world. -- / Lars Poulsen, SMTS Software Engineer CMC Rockwell lars@CMC.COM
garethh@sadss (Gareth Howell ) (06/19/91)
garethh@sadss (Gareth Howell ) writes: > Hi all, > I wonder if someone can help me with a tricky problem I am trying to come to > grips with. > > I have the (un)enviable taks of coming up with an Internet Addressing > Strategy for the UK Department of Social Security's internet (note small > 'i':-). This comprises (or soon will) 2500+ Ethernet LANs: each with > anything from 4-50 PCs, Unix application servers and gateways on them; all > interconnected using the Government Data Network (X.25 (1980)). Most of the > operational systems use OSI protocols, but there is a significant amount of > IP traffic, mainly for SNMP HUB and BRIDGE and host management on the LANs. > I have had a couple of mail replies to this query (thanks guys), but unfortunately the advice has been contradictory. On the one hand I have been told that this will work if we use OSPF as the routing protocol (provided we have variable length subnet masks). On the other hand I have been told that this will not work as all subnets of a network must be physically connected. The latter was my view as well until I read Comer's book "Internetworking with TCPIP". In this (section 16.9) he deals with the situation where a host on a non-subnetted network needs to talk to one of two hosts, each of which is on a different, unconnected, subnet of a second network. This seems to be what I am trying to deal with as well. So, the questions are, a) Is Comer section 16.9 correct? b) Must the individual subnets of a network be connected? Gareth _____________________________________________________________________________ | | | | | EMAIL: | ADDRESS: | | | ukc!sadss!garethh | DSS, Moorlands Road, Lytham St Annes, | PHONE: | | ukc!cix!garethh | Lancs, England. | +44 253 797205 | |___________________|_______________________________________|_________________|
medin@NSIPO.NASA.GOV ("Milo S. Medin", NASA ARC NSI Project Office) (06/21/91)
Gareth, if you run OSPF and the router implements variable length subnet support properly, subnets of a subnetted network do not need to be connected. We do this at NASA. It does work. It should work with any vendor who implements according to the spec. We sort of stumbled into this by accident when configuring some routers, but since then, we've found it very useful. It shouldn't be used as a way to kludge around a bad design, but there are many cases where it's useful. I suggest you ask the router vendors you are taking bids from for this support. I expect most of them will tell you they can now support this configuration. Thanks, Milo