|
|
|
The objective of the load optimization process is to move traffic off a link when the decision-maker process’s estimation of the link’s load approaches user-configured thresholds, which, if cost optimization is your primary goal, should reflect your tolerance for breaking through billing tiers.
The load optimization process lets you factor in what is usually referred to as the 95th percentile. This percentile (it may be some number other than 95) identifies the percentage of usage samples that your service provider uses to calculate your monthly bill. In this billing model, samples in the top fifth percentile are discarded, to allow for momentary spikes in usage without penalty.
The decision-maker process uses user-defined thresholds in its load optimization calculations.
But since it cannot know in advance which samples will fall into the 95th percentile, it uses multiple user-configured usage thresholds to determine when to divert data from individual links. Basing your threshold table’s total configured load on your 95th percentile numbers for each of your links during the most recent month is a good starting place for configuring a table.
As long as the total load obtained from your edge routers fits below the configured total load for a level, the decision-maker process will try to keep the load on each link below the thresholds you have configured for the link.
Also, since for each prefix the decision-maker process will only consider links that are in the extended winner set--which you define--the process will not allow performance to fall below what you have determined is acceptable. (See the section Route Assertion in Chapter 3: Adaptive Path Control.)
Note that this also means you need to balance your performance goals with your load optimization goals.
Too narrow a winner set value (which indicates a reluctance to compromise on performance) will make it more difficult to optimize for load.
Conversely, too large a winner set value will mean performance is never taken into account.
You can define multiple tables, one for each decision policy, for example, but there must also be one and only one global table.
Each table should have one column for each configured link (L1, L2, and L3 in the example).
Polling for link usage must be enabled for each of the links in the table (and only for links that are present in the table).
Each cell contains a maximum usage figure, expressed in megabits per second (Mbps), and a starting usage level, expressed as a percentage of the maximum value.
Each row of the table represents a threshold level. When usage exceeds the maximum defined for a level (90, 104, 118 and 132 Mbps in the example), the decision-maker process will use the thresholds in the next higher level.
The same total cannot be specified more than once (you can’t have two level 90s, for example), and the specified totals must equal the sum of the maximum-avoidance values across the row. If the total does not match the sum, the table will be invalid and the decision-maker process will reject it.
|
|
|