Specifying the Evaluation Interval

To avoid causing the CNA system to flap--the movement of massive blocks of prefixes on and off a link repeatedly in response to load samples--the load optimization process picks a random evaluation interval for each prefix based on criteria you specify. This prevents the decision-maker process from basing all routing decisions for every prefix on the same link-usage data points.

The evaluation interval is determined by multiplying a value by the SNMP polling interval, which is a period of time that was user-specified in the link configuration.

The decision maker will not evaluate the hot/cold state of links on a prefix for the period of time specified by this load-optimization evaluation interval. Each prefix gets assigned a different load-optimization interval, and a new interval is computed when the previous one expires.

When the interval time has elapsed, the decision-maker process will take the then-current load sample.

An interval of 3 times the polling interval, for example, means the decision-maker process will wait a span of time equal to three polling intervals before re-evaluating the status of a link. The results of the most recent evaluation will be reused whenever an evaluation cycle is skipped.

You need to specify two multipliers as arguments to the load evaluation-interval command:

• mean-cold--multiplier value used in calculation of the interval before the next evaluation if the decision maker process has determined the link to be cold for this prefix
• mean-hot--multiplier value if the link is determined to be hot for this prefix

A link is considered to be cold (not over loaded with traffic) if the decision-maker load-optimization process has determined that it can remain a candidate for the winner set.

A link is considered to be hot (currently carrying too much traffic) if the load-optimization process has determined that it cannot be a candidate for the winner set.

Use a higher value for the mean-hot argument to produce a longer interval for hot links, which gives the link more time to cool down before the next evaluation.

You also need to specify endpoints for the exponential curve from which the random interval is chosen, with the load evaluation-interval command’s min and max arguments:

• min--the minimum number of usage samples that will be skipped (if the randomly selected number is smaller than this value, the min value will be recomputed)
• max--the maximum number of usage samples that can be skipped (if the randomly selected number is greater than this value, the max value will be recomputed; this prevents long gaps in evaluation times for the link because of an interval chosen from the far extremity of the exponential curve)

The evaluation interval is calculated by multiplying the mean-cold or mean-hot value times the polling interval and generating a random number based on the result. (Polling interval is specified by the polling usage interval command; see Chapter 4: Polling Routers of the CNA Advanced Monitoring Guide).