| Network Description | Two network designs using Nokia and VRRP are presented. One simply provides firewall failover, while the other provides full HA. |
| Author | Jason Mogavero |
| Details | VRRP needs to be enabled on both of the firewalls, and address space needs
to be assigned for the backup IP address. This functionally works the same as HSRP, where
each interface has it's own IP address, and there is a third address that is used to
represent the cluster as a whole. Keep in mind that this is a primary/secondary setup,
rather than a load balancer, so the secondary firewall will not see any use unless the
primary fails. This is accomplished using a priority rating on each interface, and a delta
for each rating. In each configuration, the master will have a priority of 100 on each VRRP-enabled interface, with a delta of 25. This means that if there is link failure on any one of the monitored interfaces, the priority will be reduced by the amount of the delta, changing the master priority to 75. The backup firewall has a priority of 90, so on master failure, the priority-90 backup takes over the master state and becomes the functional firewall. This takes place in a span of milliseconds, and is seamless to the user and to the rest of the network. (remember that there is a virtual IP that is addressed as if it were the firewall for each interface) Here are some sample VRRP configurations for each firewall (pretend the |
Standard HA Firewall:
Two Nokia IP440s are used here, though you could just as easily use any model as long as
at least 3 ethernet ports are present. The external interfaces of the firewalls plug into
a switch which is connected to the internet router. (not shown in diagram) The internal
and DMZ interfaces go to separate VLANs on another switch. The DMZ VLAN is configured to
use publicly routable IP addresses, and the office VLAN configured for private network
addressing per RFC-1918. You could use private address space in the
DMZ and static NAT at the firewall, but this is no more secure than using public addresses
and adds unecessary load to the firewall.
The thing to keep in mind with this design is that's not a very good idea...you're
introducing several single points of failure, and it's here more for an example of how NOT
to implement high availability. However, being from a managed service environment, I've
seen this done countless times, and want to show by example of why this is a bad idea.

Fully redundant HA firewall and network:
The firewall configurations will be identical to the Standard HA design, but this network
eliminates all single points of failure. There are two ISPs running from two separate
backbones coming in, and the internet routers from them connected both to each firewall
and to each other, running BGP for dynamic routing. To accomodate for this on the
firewalls, you'll need two default gateways with different priorities assigned. The
routers are connected so that if the ISP link goes down, BGP will route outbound traffic
to the other router and out to the other ISP. The router connected to the ISP that had
gone down would still show as up to the firewall, so the weighted default route wouldn't
take effect here, as it could still route to the higher priority gateway.
Behind the firewall, there are two switches each segmented into 2 VLANs. The VLANs on each
switch will be configured to use the same subnet as the same VLAN on the other switch. The
DMZ runs off of VLAN1 and the office LAN from VLAN2. Interface 2 on each firewall will run
to the DMZ (VLAN1 on either switch) and Interface 3 goes to the office VLANs. You can
cross connect the switches, run a crossover between VLANs, or dual home each server in the
DMZ to each VLAN, depending on your environment. The office network, depending on it's
size, will probably sit behind a router which can
be dual homed to VLAN2 on both switches. Of course, at this point you're introducing a
single point of failure, but it's no longer cost effective to devote this much resource
into non-mission-critical systems.
Traffic between VLANs (such as management of the DMZ web servers) will need to be routed
through the firewall, so a rule will need to be in place to allow for this as well.
