Having fun with VRRP and Clustering on Vyatta

Having had some time post teaching to get my head down to some serious network building, I've been investigating building high resilience in to my hosted call centre designs.

I base my design as much as possible on Open Source technologies, since it gives me and the people I work for the biggest bang per buck for the technology deployments. There's been discussion on a number of blocks about Vyatta's claims around their router performance vs. Cisco and even some flaming of both parties - none of that here!


Vyatta has a place in any network. Why? Flexibility (well and of course cost), by utilising open source, standard hardware and pre-integrated hardware like Sangoma cards, you have a very versatile routing capability. Plus its easy to extend to to anything you can imagine. For example - for building redundancy across a number of routers to cope with failures VRRP has been developed in the IETF (based on Cisco HSRP), but failing over things like IPSec tunnels can be a bit more fun, so use clustering instead. HA cluster is already well implemented under Linux, but combining in a routing platform is genius, especially when its easy to configure.

We've been building routers and firewalls using Linux for a while (IP Tables and the Kernel can be a pretty slick firewall if you're happy to roll-your-own, and to be fair there are a few ways of doing it including other open source solutions like Smoothwall. Horses for course - Yes I like Smoothwall too!

The other main bonus of a Linux based Firewall Router is virtualisation - with huge pressure on data centres both from the power consumption and cost plus the greening up of IT, virtualising the resources of service and your routing/firewall infrastructure is very compelling. No a don't think a Vyatta pizza box could replace the layer 3 48Gpbs non-blocking switch right now - but if I'm virtualising my data centre - where's the switch? Actually there are fewer switch ports too, because each server could be running many Operating System instances with "software" ethernet bus bridged to the physical NICs on the device. We're in a position to team NICs in to an aggregate LAN connection of say 8 Gbps with 8 NICs in a team, and handle massive amounts of compute power in a single redundant chassis.

So just how easy does something like Vyatta make say VRRP:
interfaces {
ethernet eth0 {
address 172.16.0.252/24
hw-id 00:15:c5:fa:3c:1a
vrrp {
vrrp-group 100 {
priority 200
sync-group hotswap
virtual-address 172.16.0.254
}
}
}
ethernet eth1 {
address 192.168.25.252/24
hw-id 00:15:c5:fa:3c:1b
vrrp {
vrrp-group 200 {
priority 200
sync-group hotswap
virtual-address 192.168.25.254
}
}
}
loopback lo {
}
}

The configuration above creates a VRRP enabled interface (the master router), which will swap over to a slave in a pair (or more you can have more than two routers) if either of the Interfaces fail. Sweet.

So putting that kind of thing to good use - I can put a NIC from each router onto a separate switch if the switch fails connecting the master, then the backup will swing into action within a couple of seconds. (TCP connections won't see the joint). Now depending on how the servers connected to the switches are configured, combine NIC teaming on the servers, and you have a highly available solution.

Combine allow that with the flexibility of IPTables and an easy configuration you've a lot of bang for buck!