load evaluation-interval

Sets a required time interval between load samples used for load optimization.

load evaluation-interval <min> <mean-cold> <mean-hot> <max>
 
no load-evaluation
 

• min--minimum number of SNMP samples to skip before re-evaluating a link; range: 1 through 100
• mean-cold--multiplier used to calculate the random variable for cold links; range: 1 through 100
• mean-hot--multiplier used to calculate the random variable for hot links; range: 1 through 100
• max--maximum number of samples to skip; range: 1 through 100, with max – min >= 3

Must be in engine or policy configuration mode (config-engine or config-engine-policy), which requires privileged-level access.

Use once globally, or once per policy block.

min--3

mean-cold--5

mean-hot--6

max--10

The load evaluation-interval command sets multipliers which will be used by the CNA decision-maker process to spread out over time successive evaluations of the state of a link for load optimization. Forcing the decision maker process to choose a different interval at random for each prefix prevents flapping of the system (mass movements of prefixes on or off a link in unison because calculations for each prefix are based on the same data points).

The interval is expressed as a multiplier of usage samples obtained by the CNA link-monitor process.

See the polling usage on command for more information about polling.

The actual multiplier used by the decision maker process is selected at random from an exponential distribution. The multiplier is selected at random in an effort to avoid clashes in time among prefixes as the decision-maker process works through its queue of decisions to be made.

The min argument establishes a minimum amount of time--a multiple of the polling interval--to wait before re-evaluation of the link. You should not allow the decision-maker process to use consecutive load samples to evaluate a prefix. Using different load data points for different prefixes helps avoid system flapping.

The max argument sets a maximum amount of time to be skipped before re-evaluation of the link, in order to avoid selection of an interval far in the future (because of the long tail of the exponential distribution from which the random interval is selected).

Be sure the difference between min and max is at least 3.

The mean-cold argument is a multiplier used by the decision maker process to compute a random interval for re-evaluating cold links.

The mean-hot argument is a multiplier used by the decision-maker process to compute a random interval for re-evaluating hot links.

The mean-hot argument should be larger than the mean-cold argument, to give hot links more time to cool off while still paying sufficient attention to cold links.

See also the following commands:

• load min-required-samples
• load optimization
• load threshold-table

Use the no form of the command to set all arguments to their defaults.