Amazon.com, Inc.
Network routing tables generated based on pricing information

Abstract:

In a border network having a plurality of routers, routing tables for the routers can be generated using one or more weighted parameters, such as a price associated with transmission of packets over third-party routers, capacity of the border routers, and performance information (e.g., latency) of the border routers. The routing tables can be repetitively generated to proactively ensure that the border network is operating optimally. The framework described augments a control plane with Availability, Performance, and Cost (APC) telemetry collected from existing services in a cloud environment. Such telemetry can be continuously correlated in a convex optimization framework to maintain the network in its optimal state. As a result, the framework can proactively avoid potential network degradation events (e.g. congestion) and the resulting customer impact, while reducing the cost of routing traffic through upstream peers/transits.