I’m continuing the list of NMS event handling requirements.
In the last post, Handling Event Data, Part 1, I talked about correlating interface up/down events. I would also like to know if an interface is unstable. This is where the interface is going up and down on a regular enough basis that it is a hint that something needs further investigation. Any event type that has a count greater than some value over a given time interval would create an alert or would appear in a report. For example, report any interfaces that have had more than two ‘down’ events per week. I would use this information to identify and research intermittant problems and work to make the network more stable and to reduce churn in the routing and switching protocols.
Alert severity changes
An event (syslog or Trap) is received, it is processed and may generate an alert. An alert is a notification to the administrators that something needs attention (e.g., link down). When a correlating event is received, the alert may be cleared (link up), but it is useful to track that alert at a lower priority instead of clearing it. Then, if the same events recur, a counter on the alert gets incremented. It isn’t good to keep alerts around forever, so they must age out. What’s useful is a daily or weekly or monthly event/alert report. There are two ways to accomplish the tracking of repeated events. One is to record the information in the alert status. The other is to have reports that consolidate the data that is stored in the alerting and event database (note that I am not necessarily advocating that all events be kept in a database – the filesystem may be the most suitable place to store event data due to high volumes of events and the efficiency with which they can be stored).
Identifying rare events
Ideally, the common events are all identified in some way. Think of this as a filter that identifies known problems and how to handle them. Interface up/down events may be handled by generating alerts if a correcting event is not received within a certain time period, as described previously. What’s left after recognizing the common set of events are rare events. Pinnacle errors, fan failures, and power supply failures fall into this category. Such events may then be highlighted by creating a high severity alert. Simply because they are rare, they should receive extra attention, and should generate an immediate alert. The network staff can then examine whether the event is critical to the operation of the network and take appropriate action.
I’ll cover additional requirements in the next post, Syslog Summary Script.
Re-posted with Permission
NetCraftsmen would like to acknowledge Infoblox for their permission to re-post this article which originally appeared in the Applied Infrastructure blog under http://www.infoblox.com/en/communities/blogs.html