Host declarations

[Simon Hobson]

Alex Moen wrote:

>We're in a situation where we have DSLAMs scattered throughout 5 different
>towns.  The relay-agent IP addresses are all on different subnets.  Each
>DSLAM has a set of DHCP addresses assigned to the customers on that DSLAM.
>
>However, we would like to use a common pool for customers paying for
>DHCP-assigned static (some say "persistent") addresses, as the areas are
>small enough and the requests are few enough to warrant sharing between
>these towns for ease of administration and also more efficient address use.
>Normally we just put a mac address reservation for the client into the DHCP
>server and it works. However, since the relay-agent IP addresses are on
>different networks, and each system already has subnet declarations, how do
>we share a pool between them?  Is there a key phrase in the config that
>forces the server to hand out an address based on MAC address, bypassing the
>checks-and-balances that the shared-subnet statements give?

 From the basic IP networking point of view, you are wrong. It would 
be absolutely wrong for the dhcp server to allocate an address to a 
client which was not in the subnet served by the clients network - to 
do so would be to misconfigure the client so that it could not 
operate on that network.

>It seems to make sense to me that if a MAC reservation has been made, it
>should take precedence (based on the "most specific" rules that apply to
>other IP addressing things) over anything else that the server knows about.

No, see above. The client must be given an address which is actually 
usable by it.

>Does anyone have any ideas on how I can make this work?  I can put actual
>examples of subnets and VLAN info up here, but it is way complex, possibly
>more complex than is needed to solve the problem.

Well the only way I could see of doing it would be to create a large 
subnet bridges across all five 'sites', and get the relay agents to 
use an address from that subnet for those clients. What you won't be 
able to do is use a shared subnet because the server doesn't support 
the complexity required to deal with subnets A-E all sharing cable 
with subnet X, but NOT shared between each other.


I think it's a case of the software being used in situations 
(cable/dsl installations) which were simply not envisaged when the 
design was first layed out, and which are sufficiently different from 
"ethernet and serial links" networks which were the norm until only a 
few years ago.


What you could probably do is treat the whole network as one big 
network and assign addresses by option 82 (ie level 2 port id instead 
of level 3 Ip address) - but again, the current software 
implementation is not good at that and requires some messy config 
hacks to work.