Abstract: Techniques for managing computational load in real-time communications include receiving usage data on a particular node of a packet-switched network. The usage data indicates an amount of computational resources consumed on that particular node by application layer processes for a first type of data packet. It is determined whether this amount exceeds a threshold amount associated with performance degradation. If it is determined that the amount exceeds the threshold, then a message is sent to a different node on the network. The message includes overload state data that indicates the type of data packets and a reduction request. The reduction request indicates a request to reduce an amount of those type data packets that are sent to the particular node in a particular way. These techniques allow application traffic to be diverted under more specific control than approaches that stop all traffic of all data types to the node.