Designing a Load Threshold Table

The purpose of a multi-level load threshold table is to provide the decision-maker process with a cascading series of filters to assist in choosing the lowest cost alternative as the amount of traffic on your service provider links reaches billing tier boundaries.

Choosing the number of levels and the values of the thresholds is somewhat subjective, since it involves such variables as your specific network conditions, service provider billing tiers and contracted prices, and your own expectations about service provider usage.

There are some guidelines you can use as a starting point. Consider this scenario:

• three T3 lines, defined in the example CNA configuration as L1, L2 and L3, each with a maximum capacity of 45 megabits per second (maximum thresholds configured in the load threshold table are slightly lower than actual maximum capacity)
• billing tiers priced at $1,000 for up to 20 mbps on each link, $1,500 for 20-30 mbps on each link, and rates of $10,000 (L1), $50,000 (L2), and $25,000 (L3) for usage in excess of 30 mbps on each link
• each provider contract contains a 95th percentile provision (your billing will be based on 95 percent of the billing samples, allowing usage to spike into the next highest billing tier five percent of the time)

You want the CNA system to optimize traffic distribution so that L2 is the least likely to penetrate into the highest tier (since, at $50,000, it is the most expensive at that usage rage) and L1 (at $10,000) is the most likely.

The threshold table example, described in the section Load Thresholds, is designed to achieve this distribution.

The table contains four levels. The first level defines max-avoidance levels for each of the three links to be 30 mbps. This number represents the 95th percentile for each of the three links during the previous month. As long as usage stays below the 95th percentile on each link, they are given equal footing in the CNA route selection process.

When total usage exceeds 90 mbps, however, the decision-maker process will use the next level of thresholds, which identifies L1 as the link to be preferred (since it is the least costly should it break into the top billing tier). The table identifies L1 as the preferred link by assigning the link’s maximum capacity to the max-avoidance value.

NOTE: The maximum purchased capacity represents the point at which your service provider could stop handling your data. With max-avoidance thresholds configured slightly lower--44 instead of 45 in the example table--the CNA system will stop sending traffic to a link before, not when, that happens.

For similar reasons, it is advisable to configure total load on each level to be slightly less than the tier boundaries defined in your service contracts.

When total usage exceeds 104 mbps, thresholds defined in the third level in the table identifies L3 as the next favored link, since L1 will have been loaded to its maximum capacity by the thresholds in the previous level.

When total usage exceeds 118, with L1 and L3 both considered hot, traffic will then flow to L2.

The example threshold table does not define a low-usage threshold; it accepts the default for the start-avoidance percent-below-max value, which sets the low end of the threshold at 10 percent of the max-usage value. The smaller the start-avoidance percentage, the more probable the link will be considered hot for a given prefix evaluation.