[comp.dcom.sys.cisco] Serial interfaces IP number question

vaillan@ireq.hydro.qc.ca (Clement Vaillancourt) (08/14/90)

How do I number the serial links or the serial links interfaces?

We have the class B 131.195 address; we want to interconnect
3 different sites with 3 different subnets, what should I use for
serial interfaces IP numbers? Each of my 3 Cisco have one ethernet and two 
serial interfaces and we want to make a triangle with 3 Cisco's and 
3 serial links.

The machines and the Cisco ethernet interfaces on each subnet have numbers
in their own subnet.

I don't want to waist subnet because we are planning to interconnect
about 100 sites eventually.

We use a 9 bits subnet (max. of 128 subnets with 512 machines each);
	Site A = 131.195.2.x     We have about 400 machines per site
	Site B = 131.195.4.x
	Site C = 131.195.6.x

-What IP number should I use for the serial interfaces in each Cisco?

-How should I configure the Cisco's to make shure the traffic can
 reach each sites in the event of a single serial link failure?

I like this list very much!

Clem.
--
   Clement Vaillancourt,             |   Institut de Recherche d'Hydro-Quebec
   Network Manager                   |        1800 Montee Ste-Julie, Varennes
                                     |             P. Quebec, Canada, J3X 1S1
   Email: vaillan@ireq.hydro.qc.ca   |  Tel: 514-652-8238 / Fax: 514-652-8309

kwe@bu-it.bu.edu (Kent England) (08/14/90)

In article <2442@s3.ireq.hydro.qc.ca>, vaillan@ireq.hydro.qc.ca (Clement
Vaillancourt) writes:
> How do I number the serial links or the serial links interfaces?
> ...
> 
> I don't want to waist subnet because we are planning to interconnect
> about 100 sites eventually.
> 

	You could use unnumbered serial lines, but you would lose the
ability to talk about them in SNMP.  I heard that cisco supports
unnumbered serial lines "now", but don't ask me what release that is.

	Or you could try and get cisco to implement a routing
protocol that carries along subnet mask information, allowing
the subnet masks to vary from one interface to another.  OSPF comes
to mind as a candidate protocol.

	--Kent

leinwand@hpccc.HP.COM (Allan Leinwand) (08/15/90)

>/ hpccc:comp.dcom.sys.cisco / vaillan@ireq.hydro.qc.ca (Clement Vaillancourt) /  7:26 am  Aug 14, 1990 /
> 
> We have the class B 131.195 address; we want to interconnect
> 3 different sites with 3 different subnets, what should I use for
> serial interfaces IP numbers? Each of my 3 Cisco have one ethernet and two 
> serial interfaces and we want to make a triangle with 3 Cisco's and 
> 3 serial links.
> 
> I don't want to waist subnet because we are planning to interconnect
> about 100 sites eventually.
> 
> We use a 9 bits subnet (max. of 128 subnets with 512 machines each);
> 	Site A = 131.195.2.x     We have about 400 machines per site
> 	Site B = 131.195.4.x
> 	Site C = 131.195.6.x

I know this is not the answer you want to hear, but let me tell you our rule 
of thumb concerning making a subnet mask and picking a network number:

1.  estimate the total number of subnets needed.

     You did this.  With serial links and all you potential sites this number
could exceed 150.  Find the closest power of 2 which exceeds this number.  For 
your setup I get 2^8 == 256.  So, you need 8 bits for the subnet number in your
subnet mask.

2.  estimate the total number of machines per subnet.  

   Again, you did this and came up with a number of 400.  This equates to
2^9 == 512.  So, allow 9 bits in the subnet mask for the host entries.  

3.  add the two powers expressed above and find the closest network class
needed.

   In your case I find you need 9+8 == 17 bits to express your subnet numbers
and host numbers.  I would have applied for a class A network number from the 
NIC.  Class A network give you 24 bits to play with, while class B only give you
16 bits of fun.  Is this painfully obvious? :-)

Realizing that this does not help much, I guess you will have to 
eliminate IP subnets on the serial links.  I think you can do this with an
interface sub-command as of 8.1.  Of course, this means network management 
could be a bear since you cannot ping the serial interfaces or query them 
via SNMP.

> -How should I configure the Cisco's to make shure the traffic can
>  reach each sites in the event of a single serial link failure?

Make sure the IGRP metrics match (by setting the bandwidth and delay parameters
the same) on all links in the triangle.  IGRP will provide load balancing
across the two links if the IGRP metrics are identical.

Take care,

Allan Leinwand
Hewlett-Packard
Network Solutions Engineering
allan@hpcta.corp.hp.com

"Remember, no matter where you go -- there you are."
   - Buckaroo Banzai

forster@cisco.com (Jim Forster) (08/16/90)

Folks,

While it is true that you cannot ping unnumbered serial interfaces,
fortunately, the SNMP designers have designed things so that it *IS*
possible to query unnumbered interfaces using SNMP.  The SNMP packets are
aimed at any IP address of the router, and refer to interface numbers
(i.e., the 1st, or 2nd, interface).  Another effect of this is that
you can use SNMP to query about non-IP speaking interfaces.

 -- Jim Forster
    cisco Systems


PS: It is not possible to netboot a router over an unnumbered serial line.

BILLW@mathom.cisco.com (WilliamChops Westfield) (08/16/90)

> We have the class B 131.195 address; we want to interconnect
> 3 different sites with 3 different subnets, what should I use for
> serial interfaces IP numbers? Each of my 3 Cisco have one ethernet and two 
> serial interfaces and we want to make a triangle with 3 Cisco's and 
> 3 serial links.
> 
> I don't want to waist subnet because we are planning to interconnect
> about 100 sites eventually.
> 
> We use a 9 bits subnet (max. of 128 subnets with 512 machines each);
> 	Site A = 131.195.2.x     We have about 400 machines per site
> 	Site B = 131.195.4.x
> 	Site C = 131.195.6.x

Well, I really like the idea I heard someone suggest a while ago on some
other list - assign the subnet bits high-bit first, and the host addresses
low bits first.  This allows you to easilly expand the subnet field as
the number of nets you are actually using goes up.

For example, consider a class B address N.N.0.0.  If you only have 3
networks now, you can get by with a 3 bit subnet mask (255.255.224.0),
and assign 6 networks:

	N.N.128.0
	N.N.64.0
	N.N.192.0
	N.N.32.0
	N.N.96.0
	N.N.160.0

If you need to add more networks, you change the netmask to 4 bits
(255.255.240.0), freeing up an additional 8 networks, but all the old
network numbers are still valid.

This way, you only run into a problem if you actually get too many
networks with too many hosts on each one.

With the 8.1 software, you need not allocate a subnet to serial links,
which may help in your case.  Also, if only a few of the networks have
400 hosts, you can set up the ciscos to put two subnets worth of hosts
on a single cable.

Bill Westfield
cisco Systems.
-------

kwe@bu-it.bu.edu (Kent England) (08/17/90)

In article <24823@boulder.Colorado.EDU>, 
BILLW@mathom.cisco.com (WilliamChops Westfield) writes:
> 
> Well, I really like the idea I heard someone suggest a while ago on some
> other list - assign the subnet bits high-bit first, and the host addresses
> low bits first.  This allows you to easilly expand the subnet field as
> the number of nets you are actually using goes up.
> 
	How do you suggest dealing with the node configurations?  Should
one be constantly changing the subnet mask?  Use proxy arp forever?  Wait
for dynamic host configuration?

	I wish it was easier to change these sorts of things as we
go along.

	--Kent

RAF@CU.NIH.GOV (Roger Fajman) (08/21/90)

> Well, I really like the idea I heard someone suggest a while ago on some
> other list - assign the subnet bits high-bit first, and the host addresses
> low bits first.  This allows you to easilly expand the subnet field as
> the number of nets you are actually using goes up.

We did that, but it didn't work out very well in practice.  The problem
is that the subnet mask exists in so many places (routers and hosts)
that people get very upset when you want to change it.  So we finally
settled on a 7 bit subnet mask and got a second class B address to use
when we run out of subnets.  That probably will be next year sometime.
We will have to assign a subnet address from each net number to our
backbone networks in order to keep the subnets of each net contiguous.

leinwand@hpccc.HP.COM (Allan Leinwand) (08/22/90)

> / hpccc:comp.dcom.sys.cisco / RAF@CU.NIH.GOV (Roger Fajman) /  9:23 pm  Aug 20, 1990 /
> We did that, but it didn't work out very well in practice.  The problem
> is that the subnet mask exists in so many places (routers and hosts)
> that people get very upset when you want to change it.  So we finally
> settled on a 7 bit subnet mask and got a second class B address to use
> when we run out of subnets.  That probably will be next year sometime.
> We will have to assign a subnet address from each net number to our
> backbone networks in order to keep the subnets of each net contiguous.
> ----------

   Yes, changing a subnet mask on every system within a network can be
very difficult.  On our network, even changing a subnet mask on all the routers
would be quite a task.

   I still think the best bet is to plan the network correctly and get the
appropriate network number assigned in the beginning.

Thanks,

Allan
allan@hpcta.corp.hp.com
Hewlett-Packard
Network Solutions Engineering

kre@munnari.oz.au (Robert Elz) (08/22/90)

    Date:        Mon, 20 Aug 90  21:13:27 EDT
    From:        "Roger Fajman" <RAF@CU.NIH.GOV>
    Message-ID:  <9008210111.AA08964@alw.nih.gov>

    We did that, but it didn't work out very well in practice.  The problem
    is that the subnet mask exists in so many places (routers and hosts)
    that people get very upset when you want to change it.

Obviously with a scheme like this you don't want to have to
change all the netmasks every time that you add a few more
subnets - the sensible approach is to numbre subnets from the
high order bits, hosts form the low order bits, and set the
netmask at some appropriate place in between (based on expected
usage of the number space).

The mask would only ever need to be altered if the anticipated
usage of the number space turned out to be incorrect.

    So we finally settled on a 7 bit subnet mask and got a
    second class B address to use when we run out of subnets.

If you have > 254 hosts on any nets, then it looks (currently)
as if you made a wise choice, which probably means that you
made it with sufficient useful information.  Sites starting
out with their first 5 hosts or so, on 3 or 4 subnets are
rarely able to predict that accurately.

If your subnet space holds out until your routers implement
variable width masks, then most of your subnets will need to
have more than 254 hosts before you would need a second net
number.

Personally, I'm hoping that variable width subnet support
will become mandatory in the forthcoming router requirements
RFC.

kre