Each SR OS node has a receiving queue per Gx application (ESM, UM, AA). Each queue can hold 10,000 messages. While the queue is in the overloaded state, the SR OS node replies to every new RAR message with the RAA (ACK) immediately followed by a CCR-U message containing the error-message with the description ‛Overload’. This can be considered as explicit signaling toward the PCRF notifying it of the condition on the SR OS node.
If the messages in the overwhelmed queue do not require sending an answer (in case that the overwhelmed queue contains CCA-I/U messages), the TCP window fills up, TCP ACKs are not sent and consequently this IS an implicit notification to the PCRF to slow down.
If the SR OS node receives a response from an overloaded PCRF (Result-Code = DIAMETER_TOO_BUSY), the SR OS node timeouts (tx-timer) the originally sent message. After the message is timed out, the configuration settings (on-failure) determines whether to trigger the peer-failover procedure or not (Peer-failover based on DIAMETER_TOO_BUSY Result-Code is recommended in RFC6733, §7.1.3.