lapyun@smosjc.UUCP (Lap Yun Yau) (04/24/91)
We just registered our Internet address and are now planning for reorganizing
of all networks within the company.
The following scheme shows the relationship of nets and subnets:
====+====================+=============== Backbone Network
| |
|Dept A |Dept B
=+==+==+==+= ==+==+==+==+==o==o=== Departmental Networks
| | | | | | | | (=o= node)
| | | =+==== | | | =+=====Net B.1 Dept Subnet 1
| | =+===== | | =+=====Net B.2 Dept Subnet 2
| =+===== | =+=====Net B.3 Dept Subnet 3
=+===== =+=====Net B.4 Dept Subnet 4
------------------------------------------------------------------------------
We want more than one backbone networks, with each backbone net connect couple
departments while each of these departments may have several subnets.
How should I setup the subnet addresses? Let's say we have a Class B address,
XX.YY., and we have two bytes (the third and the fourth octects) to play with.
------------------------------------------------------------------------------
Method 1:
Backbones - XX.YY.30, XX.YY.32, ..., XX.YY.38
Dept Nets - XX.YY.40, XX.YY.60, ..., XX.YY.240
Dept Subnets - XX.YY.41, XX.YY.42, ..., XX.YY.49 for dept net XX.YY.40
- XX.YY.61, XX.YY.62, ..., XX.YY.69 for dept net XX.YY.60
- ...
Is this setup a true subnetting? Our concerns are some software like Interleaf
using network license requires license server runs on, say, dept B net and
serves all subnets B.1, B.2, etc. At present, we have only backbone and dept
nets hook to it and each net has it own class B address. When Interleaf
license server runs on, say, dept A's net, nodes in other dept's net cannot
run Interleaf. Does anybody knows if the method 1 setting will work?
------------------------------------------------------------------------------
Method 2:
Let's play with the 3rd and 4th octects.
3rd octect 4th octect
+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+
| | | | | | | | | | | | | | | | | |
+---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+
\ / \ / \ / \ /
backbone dept nets dept subnets nodes
Then we can have 4 backbones, each backbone can have 16 dept nets, each
dept net can have 8 subnets, and each subnets can have 126 (128 minus all 0s
and all 1s) nodes. By looking at books and references on Internet Addressing,
it seems to us that it should work and really follow the standard way for
subnetting. The subnet mask for every networks is FFFFFF80. Is it right?
Are we missing anything? Can some network gurus confirm that or give us
some insight as to how to do it the right way?
If we follow this scheme, what should be the addresses for nodes like routers,
brideges, workstations, and computers attached directly to backbone net,
or dept nets.
================================================================= Super Backbone | | | |
| | | ========+======== Backbone 1
| | ========+======================== Backbone 2
| ========+======================================== Backbone 3
==+==+==+==+==+==+==+==+==+==+==+==+==+==+==+==+================= Backbone 4
| | | | | | | | | | | | | | | |
| . . . . . . . . . . . . . . =+============ Dept net 1
| . . . . . . . . . . . . . . ...
| ...
=++==+===+===+===+===+===+===+============================== Dept net 16
| | | | | | | |
| . . . . . . ==+=================== Subnet 1
| . . . . . . ...
| ...
=+=============================================== Subnet 8
****************************************************************************
uunet!smosjc!lapyuntengi@princeton.edu (Christopher Tengi) (04/25/91)
In article <642@smosjc.UUCP>, lapyun@smosjc.UUCP (Lap Yun Yau) writes: |> We just registered our Internet address and are now planning for reorganizing |> of all networks within the company. |> [...stuff deleted...] |> |> Method 2: |> |> Let's play with the 3rd and 4th octects. |> 3rd octect 4th octect |> +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ |> | | | | | | | | | | | | | | | | | | |> +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ |> \ / \ / \ / \ / |> backbone dept nets dept subnets nodes |> |> Then we can have 4 backbones, each backbone can have 16 dept nets, each |> dept net can have 8 subnets, and each subnets can have 126 (128 minus all 0s |> and all 1s) nodes. By looking at books and references on Internet Addressing, |> it seems to us that it should work and really follow the standard way for |> subnetting. The subnet mask for every networks is FFFFFF80. Is it right? |> Are we missing anything? Can some network gurus confirm that or give us |> some insight as to how to do it the right way? |> It looks like you are assuming that there must be some sort of hierarchy in a subnet addressing scheme. This is not the case. If you are going to use a single, fixed-size subnet mask (which you pretty much need to do, given most of today's IP implementations), you first need to determine how many subnets you need and how many hosts you expect to be on each subnet. With that information in hand, you can then determine how many bits you need for hosts and subnets. If it turns out that you need less than 8 bits for either side, I would suggest you start with a subnet mask of 255.255.255.0 (0xffffff00). This is the easiest for most folks to comprehend when you start talking about subnet xxx. The xxx would always be the third octet of an address in dotted decimal notation (assuming a class B address space). I would also like to suggest a method of subnet numbering that will give you some flexibility if you should find that you guessed wrong on the number of subnets/number of hosts. Start your host numbering on each subnet starting with the lowest binary bits and "counting up," yielding a sequence of: 1,2,3,4.... For the subnet numbers, start at the highest binary bit and "count down." If you were giving an entire octet (or at least the upper bits of one) to the subnet part, the sequence would be: 128,64,192,32,160,96.... This scheme will leave you with a "hole" in the middle, so that if you need to shift the mask, you will have a minimum of host renumbering to do. Let me use your first diagram to show you what I am suggesting: ====+===========================+=============== Backbone Network (x.y.128) | | |Dept A (x.y.64) |Dept B (x.y.192) =+==+==+==+= ==+==+==+==+==o==o=== Departmental Networks | | | | | | | | (=o= node) | | | =+====Net A.1 | | | =+=====Net B.1 Dept Subnet 1 | | =+=====Net A.2 | | =+=====Net B.2 Dept Subnet 2 | =+=====Net A.3 | =+=====Net B.3 Dept Subnet 3 =+=====Net A.4 =+=====Net B.4 Dept Subnet 4 Depending on the order you assigned the subnets, you might wind up with: Net Subnet Addr === =========== A.1 x.y.32 A.2 x.y.160 A.3 x.y.96 A.4 x.y.224 B.1 x.y.16 B.2 x.y.144 B.3 x.y.80 B.4 x.y.208 We are doing something very similar to this at Princeton, and have had no problems whatsoever. However, we are not running any 'network license servers' so I can't say that they will or will not work as you expect. I would hope that they were designed with subnetting in mind and would thus allow a server on any subnet in the above scheme to provide a license to any client on any other subnet. As for suggestions on how to assign IP addresses.... In addition to the method described above, we have used the convention that the gateway "to" each subnet gets an address of x.y.subnet.1. This assumes that subnet 128 is the root of a tree. In the above example, the router for dept. A would have addresses x.y.128.a and x.y.64.1. Naturally, this only counts for the first router into any particular subnet. Since we generally only have a single path for each, this has worked out OK for us. You always know the address to ping for the "downstream" side of a subnet gateway. Following this logic, you might be tempted to assign the address on the "upstream" side based on the downstream subnet number. I would caution against this, as it might impact your ability to shift the subnet mask later, if you needed to. As a final note, please realize that there is nothing preventing you from having redundant 'backbones' in your network. You may wish to have more than one path tying the department 'top level' routers together. All you need to do is make sure your routers know how to fall back to one if the other one goes dead. I hope this helps, /Chris ==========----------==========---------+---------==========----------========== UUCP: ...princeton!tengi VOICEnet: 609-258-6799 INTERNET: tengi@princeton.edu FAX: 609-258-3943 BITNET: TENGI@PUCC