Why BGP is not enough? – Infographic

It’s common knowledge that Border Gateway Protocol has no ability to make performance-based routing decisions and often routes traffic through paths that are congested or affected by routing anomalies. Noction IRP overcomes these drawbacks and is currently deployed in hundreds of networks around the world. Collecting and analyzing petabytes of data across all IRP instances worldwide offers us valuable data regarding overall Internet performance.

Noction IRP detects better alternative paths for traffic routing as opposed to the default BGP routing decisions. By gathering data across all our instances we retrieve interesting statistics about how BGP performs and learn the real numbers behind BGP’s ignorance of critical network performance indicators. Today we would like to share them with you. We collected data from both Service Providers networks and Enterprise networks to distinguish the results across these two types of networks as they have different metrics for consideration.

Firstly we identified the average volume of traffic for which IRP was able to find a better alternate path and we found that 36% of the traffic transiting a Service Provider network is routed via a suboptimal path by BGP. For an Enterprise network, this value equals 21%. These figures show that there is definitely a large portion of traffic that is not properly routed by BGP, leading to poor service delivery.

Next we evaluated the improvement in latency after making performance-based route selections with IRP. The average latency in a Service Provider network for the prefixes that IRP found a better path, came out to 247 ms. By executing performance probing across all the available providers and rerouting the traffic to the ones with the best indicators, IRP managed to drop latency on average by 63 ms. Therefore, we can conclude that BGP routing decisions result in a 26% reduction in latency for the analyzed prefixes in a Service Provider network and 31% in an Enterprise network.

We have also evaluated the loss drop after rerouting traffic through better-performing paths than the default ones. The average packet loss in a Service Provider network, for the prefixes that IRP found a better path, came out at 43%. IRP was able to reduce the loss to 7% on average. This considerable drop in packet loss implies that BGP ignorance of performance metrics is responsible for 84% of the packet loss for the analyzed prefixes in a Service Provider network and for 89% in Enterprise networks.

Some BGP routing decisions were driving traffic towards complete blackout paths. Overall IRP makes about 9 million route updates per month based on packet loss. 9% of these updates were made to avoid a full blackout that would be encountered if the traffic was sent through the default path announced by BGP.

To conclude, automating route optimization at networks’ edges based on critical performance metrics can considerably improve the overall quality of data delivery and contribute to the self-repairing capability of the entire Internet. According to Cisco Visual Networking Index the Daily total Internet traffic equals to 3.2 M TB. Currently, Noction optimizes over 35,000 TB meaning that 1.1% of total Internet traffic is optimized by IRP every second.
Below you can find an infographic that aggregates the data presented above.