Understanding Network Elements
A network element is a load-balanced group of RADIUS servers providing policy management for the Service Control Gateway.
Network elements are specified in the AAA profile that is applied to a policy and charging enforcement function (PCEF) profile. A subscriber is assigned to a PCEF profile.
Load Balancing Within Network Elements
The Service Control Gateway distributes requests to RADIUS servers across the servers in the network element.
Within a network element, a RADIUS server can be assigned a priority of 1 through 16, with 1 being the highest priority. You can have multiple servers with the same priority in a network element. All access requests are load balanced among the highest priority servers. If all the servers with the highest priority in the network element fail, then requests are load balanced among servers with the next highest priority level.
Dead Server Detection
To determine whether a RADIUS server in a network element has failed, the Service Control Gateway keeps track of how often requests sent to a server time out and must be retransmitted. If the number of times that requests need to be retransmitted reaches a configured limit within a configured time interval, the Service Control Gateway marks the server as dead and starts sending requests to the next available server in the network element with the same priority.
At the same time, the Service Control Gateway starts a timer for the RADIUS server. After this timer expires, the Service Control Gateway marks the dead server as alive again, and includes it in the rotation for sending RADIUS messages.
Maximum Pending Requests for a Network Element
You can configure the maximum number of requests that can be queued to the network element. When the pending request queue is full, any additional requests are dropped.
You can also configure a high and a low watermark that are percentages
of the maximum number of requests that can be queued. If the number
of pending requests reaches this high watermark, a
flow control on message is generated. When the number
of pending requests then falls below the low watermark, a
flow control off message is generated.