[Kea-users] i do not understand the concept of shared networks

[Cathy Almond]

On 10/03/2020 13:14, Giso Kegel wrote:
> Hi Folks,
> 
> I work with kea 1.6.1
> 
> I really do not understand the concept or at leased the documentation
> regarding shared-networks.
> Do i need shared-networks or not?
> I do not have the problem that I have networks that grew out of there
> original definition.
> 
> https://kea.readthedocs.io/en/latest/arm/dhcp4-srv.html
> 
> lets say i have 50+ subnet4 that will ask the kea dhcp server over dhcp
> relayhosts (a swtich).
> 
> All these subnet4 will hold only server that have a IP-reservation.
> 
> Do i need multi shard networks?
> Can or do i have to put these 50+ networks in multi shared networks?
> Do i have to use different interfaces  for each shared network?
> 
> I would be very thankful for some enlightenment.
> 
> Best
> Giso

What are shared-networks (and do you need to use them or not)?

They're a way to tell the DHCP server that all of the subnets specified
inside that one shared-network can be considered equal - IP routing of
clients should not be broken if you give any client requesting an
address, any IP address from any of the included subnets.

That's the very simple case - and it's pretty clear that if you ran out
of addresses for a specific site and needed to add a new subnet for it,
running over the same wire and sharing the same local relay, that this
is a primary use case for shared-networks - it doesn't matter which
subnet provides the address to a client, any/all should work just fine.

But there is more to this.  Clients in a shared-network are those who
all reach the DHCP servers via the same route - be it a local interface,
or, more likely, the same relay.  Consider the case of cable modem
provisioning where different types of 'client' will request addresses -
all via the same relays, but with the need to distinguish between them
and allocate them to different subnets.  Typically it will be possible
to separate clients by means of Classification and then use this to
restrict clients within a shared-network to the subnets that they should
be using.  This is the second, and slightly more complex use case.

The basic principles are therefore:
- all the clients are reaching the DHCP server via the same local relays
(or are local to it)
- unless classification is added, it's OK for a client to get an address
from any subnet within the shared-network.

(And in all of the above, I have pretended that host reservations don't
exist.  They can and will of course change how addresses are allocated
to clients, but having got the basics on shared-networks and
classification, now reading the documentation on what's different when
there are also host reservations to consider should be a bit easier...)

Cathy