If all you want to achieve is separate users from each other you can try setting Private VLAN or port isolation. This way each port can only communicate with gateway (uplink port). With such setup it does not matter whether DHCP server is directly on switch or external, but for flexibility reasons it is better to be external and just configure dhcp relay on switch to forward all requests to dhcp server.
Also if switch support "gleaning" to see DHCP transactions or DHCP lease query (RFC4388) it can even block all attempts of users to use unathorized IP addresses. For example on Mikrotik equipment you can use embedded DHCP server to put all DHCP clients into ARP tables directly and by turnning off arp responses it can force clients to only use DHCP assigned addresses.
This is theory. In practice it all depends on the actual hardware and software involved. By giving more info we can give you better answers with examples.
I think you misunderstood how a router process a packet, thus coming with a solution that is not at all appropriate for your needs.
Why?
Let say computer A has the following configuration:
- mac address
00:53:BA:12:17:19
- IP address
192.168.0.7
- subnet mask of
255.255.255.0
- default gateway
192.168.0.1
A send a packet to the internet host www.example.com which has IP address 203.0.113.5.
The packet has the following characteristics:
- source IP address : 192.168.0.7
- destination IP address: 203.0.113.5
It compare (in binary) its subnet mask with the destination IP address and find that the destination is not on the local subnet, so it will send the packet to its default gateway, 192.168.0.1
It lookup in its ARP table and if needed perform an ARP request to find the mac address of the host which hold the 192.168.0.1 IP address.
It finds 00:53:00:17:a7:b3
Then it builds a frame with the following characteristics:
- source mac address: 00:53:BA:12:17:19
- destination mac address: 00:53:00:17:a7:b3
inside this ethernet frame the IP packet is encapsulted, and it still has:
- source IP address : 192.168.0.7
- destination IP address: 203.0.113.5
As you can seed the destination IP address is NOT the gateway.
So the router receive this frame, strip the Ethernet header and lookup the packet to perform a routing decision.
The basic of routing is that the routing decision is made solely on the destination IP address, 203.0.113.5
The router then look in its routing table, find a route for 203.0.113.5 and send the packet through the associated interface (performing NAT if configured which is required here).
As you can see, the IP address of the gateway that was used has no role in the routing decision. And, more importantly, the router does not even know what was this IP address. It only know on which interface the frame arrived
Ok so, why not configuring 2 different gateways on two different interfaces. Well you can't, not on a Cisco router. You cannot have two overlapping networks on two different layer 3 interfaces. Otherwise the router could not decide on which interface it must send a packet for this network.
This is why your dual gateway cannot work.
But more importantly, it's not required to achieve your goal.
What could work?
Now if you want the router to take a different routing decision based on the sender, it is possible. It's called policy based routing (PBR)
PBR allow you to configure different routing table on the router, and perform routing decision on different criteria.
The most common (and easy to configure) criteria are the source IP address and destination IP address.
Note that you can specify the outgoing interface rather than the next-hop IP, which is handy for a outgoing interface configured by DHCP.
So what you have to do (if I understood correctly what you want), is to:
- set a group of computers with specific IP address pool (fixed IP, DHCP reservations)
- set another group of computer with a second IP address pool
- write a route map that will set the destination IP or outgoing interface for each pool
- activate PBR on the incoming interface (the one that has the LAN gateway)
To manually change the outgoing interface for some computer in case one link fail, you just have to alter the route-map, which is a matter of minutes.
You can have 4 pools for example:
- computers that will always use ISP 1, and never fail-over to ISP2
- computers that will always use ISP 2, and never fail-over to ISP1
- computer that will use ISP 1 if available, and manually fail-over to ISP2 if needed
- computer that will use ISP 2 if available, and manually fail-over to ISP1 if needed
Best Answer
None. Each client uses its default gateway. You could also add entries to their routing tables, making them use both gateways depending on the destination.