7.6 Device-to-Manager Communication

In addition to communication initiated by the manager or operator via SNMP, CLI, or HTTP, agent-to-manager communication is another important part of the management model. For example, a manager may be alerted when a port goes down or when traffic on a port exceeds a threshold. This unsolicited alert is sent on the same communication channel as the manager-to-agent interaction. SNMP uses a mechanism called a trap. With a CLI, the alert is printed on the user console.

The type of information the operator requires, the filters on the alerts, action taken on the alerts, and so forth are all areas that merit additional discussion. However, the significant issues related to an embedded communications device are:

1. The protocol task uses the same queuing mechanism for traps or alerts as it does for get or set responses. The protocol task does not send traps directly to the manager. Rather, it sends alerts to each of the agents (CLI, SNMP, CORBA, etc.). The agents will determine when to pass along the alert.

2. Agent design should restrict the number of traps to space them such that they are more manageable. One technique is to avoid sending a trap to a manager if you had sent it the same trap, say, 5 seconds ago. This is known as throttling or rate limiting

3. The alerts/traps should be grouped into critical and non-critical at the protocol or system task level. The design should ensure that critical alerts are sent before thresholds are exceeded rather than after depletion. This allows the manager to gracefully shut down the system.

This device may be one of several systems which are sending alerts, so be sure to use alerts and traps sparingly and only for critical events