Predictive overlay network architecture
The predictive overlay network architecture of the present invention improves the performance of applications distributing digital content among nodes of an underlying network such as the Internet by establishing and reconfiguring overlay network topologies over which associated content items are distributed. The present invention addresses not only frequently changing network congestion, but also interdependencies among nodes and links of prospective overlay network topologies. The present invention provides a prediction engine that monitors metrics and predicts the relay capacity of individual nodes and links (as well as demand of destination nodes) over time to reflect the extent to which the relaying of content among the nodes of an overlay network will be impacted by (current or future) underlying network congestion. The present invention further provides a topology selector that addresses node and link interdependencies while redistributing excess capacity to determine an overlay network topology that satisfies application-specific performance criteria.
1. A method of reconfiguring an overlay network in a manner intended to increase stability and reduce service interruptions, the overlay network including a plurality of nodes interconnected by links, the method comprising the steps of:
generating session duration predictions for prospective parent nodes to identify nodes that are likely remain part of the overlay network;
predicting the node-relaying capacity of prospective parent nodes;
identifying one or more supernodes, wherein supernodes are identified as nodes that have high node-relaying capacity and that they are likely to remain part of the overlay network based on their predicted session duration;
generating one or more prospective overlay network topologies in which supernodes are placed at higher levels of the overlay network topology; and
reconfiguring at least a portion of the overlay network in order to implement one of the one or more prospective overlay network topologies.
2. The method of claim 1 wherein nodes having higher node-relaying capacity encompass those nodes in the top 20% of predicted node-relaying capacity, and nodes having lower node-relaying capacity encompass those nodes in the lower 80% of predicted node-relaying capacity.
3. The method of claim 1 wherein the overlay network is a non-tree-based topology, and wherein at least a portion of the plurality of nodes are capable of receiving content segments from multiple parent nodes simultaneously.
4. The method of claim 1 further comprising generating a plurality of metrics corresponding to the plurality of nodes.
5. The method of claim 4 further comprising categorizing the plurality of nodes into a plurality of clusters based on one or more metrics.
6. The method of claim 5 wherein at least a portion of the plurality of metrics have associated timestamps, and wherein the one or more metrics for categorizing the plurality of nodes include the metric of time.
7. An adaptive topology server adapted to reconfigure an overlay network in a manner intended to increase stability and reduce service interruptions, the overlay network including a plurality of nodes interconnected by links, the system comprising:
a prediction engine that generates session duration predictions for prospective parent nodes to identify nodes that are likely remain part of the overlay network, and predicts the node-relaying capacity of prospective parent nodes; and
a topology selector that identifies supernodes, wherein supernodes are identified as nodes that have high node-relaying capacity and are likely to remain part of the overlay network based on their predicted session duration, and
wherein the topology selector generates one or more prospective overlay network topologies in which supernodes are placed at higher levels of the overlay network topology, and
wherein at least one of the one or more prospective overlay network topologies generated by the topology selector are used to reconfigure at least a portion of the overlay network.
8. The adaptive topology server of claim 7 wherein at least one of the one or more prospective overlay network topologies is a non-tree-based topology, and wherein at least a portion of the plurality of nodes are capable of receiving content segments from multiple parent nodes simultaneously.
9. The adaptive topology server of claim 7 further comprising a metrics processor that generates a plurality of metrics corresponding to the plurality of nodes.
10. The adaptive topology server of claim 9 wherein the topology selector categorizes the plurality of nodes into a plurality of clusters based on one or more metrics.
11. The adaptive topology server of claim 10 wherein at least a portion of the plurality of metrics have associated timestamps, and wherein the one or more metrics for categorizing the plurality of nodes include the metric of time.
12. The adaptive topology server of claim 7 wherein nodes having higher node-relaying capacity encompass those nodes in the top 20% of predicted node-relaying capacity, and nodes having lower node-relaying capacity encompass those nodes in the lower 80% of predicted node-relaying capacity.