Surely a non-topic? Isn’t traffic engineering a simple case of throwing internet traffic towards the lowest cost destination? This may be how many Internet Networks approach traffic engineering in their earliest days, but how far can this strategy scale?
Most of the time, the strategy breaks when a failure event on the network poses the question “could we have done better?” A failure event could be an outage that leads to congestion, loss of service, or unnecessarily high latency to a particular destination. Most of these have cost (sending traffic “around the houses”) or revenue (customers getting fed up and leaving) implications, but for most networks in the world some careful planning significantly reduces the impact of failure at their Internet edge.
Managing traffic engineering projects can be boiled down to a simple checklist:
- Establish the ‘direction’ of your traffic flow (mainly inbound, mainly outbound, or balanced)
- Collect data that you can trust which helps you to identify which remote ASNs originate or sink your top traffic flows
- Obtain quick wins by configuring deterministic routing with your top ‘n’ flows
- Improve mutual long term performance for both networks by interconnecting closely to traffic origination point.
- Measure again, review again, measure again, review again.
The techniques that enable this are explain in more detail in this presentation.