Point-to-point RPC and Web Services. In many of the cases where systems need highly reliable and low latency point-to-point communications, such as for Web Services, we are able to utilize the Akamai high-performance overlay described in Section 6.1) to improve reliability and performance, even in the face of network problems. Dynamic configuration updates.Many Akamai system components need to receive frequent configuration updates with low latency, sometimes as frequently as every few minutes. One example of this is the customer metadata configuration files used to configure the edge server platform, as described in Section 7.1. Key design goals here include version consistency across the network (including graceful handling of connectivity issues and machines that can fail and restart at anytime, reliability and scalability of the system, and a mechanism for ensuring that propagated changes do not adversely affect the network. Our approach is to publish the data to a set of highly-available storage servers that use quorum-based replication to accept an update. To achieve scale and low latency in distribution, updates are then propagated throughout the network using Akamai‘s own content delivery services. Finally, configuration rollouts are automatically phased, with health checks performed at each step, to protect the network. Further details, including the vector-exchange-based acceptance algorithm and the index merging recovery mechanism, are covered in [39].
