Akamai customizes open source congestion control algorithm

Akamai is using a congestion control algorithm that was originally developed by Google to optimize streaming video traffic on its content delivery network (CDN) platform. The algorithm is called the Bottleneck Bandwidth and RTT (BBR).

Internet traffic rides on an underlying transport protocol, the most popular of which is TCP. As CDN traffic is delivered into last mile networks, it often hits a lot of interference that increases degradation. Congestion control algorithms are used to combat that degradation. Alex Balford, senior product marketing manager at Akamai, said Akamai uses about five different congestion control algorithms that work with TCP. Some of these algorithms are proprietary and some are open source.

BBR is the newest of Akamai’s congestion control algorithms. It was developed by Google, which has open-sourced it for the larger community. “This algorithm matches the bandwidth requirements with the send rate of packets,” said Balford. “You want to right-size that to not send too fast or too slow.” He said Akamai tested BBR for more than a year in order to understand how it would operate within its global platform. Akamai tested it in its labs and then with different customers and different variables. “We didn’t want to blindly impact our customers,” said Balford.

After more than a year of testing, in November the company rolled out a version of BBR that is optimized specifically for the Akamai network.

The BBR rollout applies to global traffic between the Akamai Edge network and client devices for Akamai’s Adaptive Media Delivery and Download Delivery products. Adaptive Media Delivery is for the delivery of music and video streaming. Download Delivery deals with large software or game downloads.

Balford said that Akamai has a deep understanding of global networks because it works with 1600 of them around the world. The CDN provider also has a lot of data related to internet devices. “We have the ability to harness all this data we gather on a daily basis. And we use machine learning to automatically deliver the best experience and make decisions in real time.” He said that Akamai discovered through machine learning that BBR was often the best congestion control algorithm to use for a particular end user in a given moment.

BBR is now serving more than 80% of end user connections on the Adaptive Media Delivery and Download Delivery products.