BGP Route Prioritization

Use Cases for BGP Route Prioritization

Table 1 shows the types of routes that would benefit from route prioritization and some notes about why they would benefit from it. Examples of those types of routes are also included. Prioritizing these routes within a given large-scale environment can help routers to react more quickly to important route changes.

Properties of BGP Route Prioritization

BGP route prioritization in Junos OS is implemented using a set of 17 prioritized (numbered) output queues that are serviced by a user-configurable token mechanism. This section describes the prioritized output queues, the operation of the token system, and assignment of routes to queues.

• Prioritized Output Queues
• Assignment of Routes to Queues
• Work Token Mechanism

Understanding Queue Priority and Fairness

The scheme used by BGP route prioritization focuses on two elements: fairness and priority:

• Fairness means that when there is work to do in any given queue, other queues are guaranteed to get some work done at some point. How much work each queue is permitted to get done is determined by the number of tokens assigned to each priority.

• Priority means that when there is competing work and fairness has been ensured, to always choose the more important work.

For example, presume three classes of priority: low, medium, and high. These could be assigned to queues 1, 2, and 3, respectively. Alternatively, they could be assigned to queues 3, 6, and 9. For fairness, if the decision is that high priority gets 50% of the available work, medium gets 35%, and low gets the remaining 15%, tokens can be assigned as 50 to high, 35 to medium, and 15 to low. Alternatively, tokens can be assigned as 5 to high, 4 to medium, and 2 to low. You can assign any of the 17 queues any value between 1 and 100. The ratio of the number of tokens in a single queue to the total number of tokens in all queues gives the percentage of work that will be done in each queue.

Priority is most important when work appears in a queue while tokens are in the process of being spent in another queue by the work scheduler. Table 3 shows the starting point for an example of this.

If we assume that the work scheduler is processing the medium queue (queue number 6) and has spent 20 tokens, then there are 15 tokens left to be spent on the remaining entries in the medium queue and 15 tokens left to be spent in the low priority queue. If 5 entries arrive in the expedited queue prior to the next run of the work scheduler, those 5 entries will be sent first because there are still 50 tokens left in the expedited queue.

Queue Servicing Procedure

The queue servicing procedure operates per-BGP peer group with each group maintaining its own token buckets.

• Token buckets for each priority start full either at the configured number of tokens or at the default of 1.

• Each time a route entry is pulled from a queue to start a BGP update, a token is subtracted from that queue.

• While the expedited queue has tokens, every other queue entry is drawn from the expedited queue, subject to the route packing rules.

• Entries are taken from the queue that has the highest priority. This means that if entries are added to a higher priority queue between runs of the queue servicing mechanism, and there are tokens available in that higher priority queue, the new entries in the higher priority queue are sent first, thus preempting entries in lower priority queues. If the higher priority queue has no work tokens available when the new entries arrive, the new entries are not sent until after the next token refresh.

• Tokens are refreshed after all priority queues have been serviced (there are no entries remaining in any queue) or when all tokens are exhausted.