[Contents] [Prev] [Next] [Index] [Report an Error]

Allocation of Leftover Bandwidth

The allocation of leftover bandwidth is a complex topic. It is difficult to predict and to test, because the behavior of the software varies depending on the traffic mix.

If a queue receives offered loads in excess of the queue’s bandwidth allocation, the queue has negative bandwidth credit, and receives a share of any available leftover bandwidth. Negative bandwidth credit means the queue has used up its allocated bandwidth. If a queue’s bandwidth credit is positive, meaning it is not receiving offered loads in excess of its bandwidth configuration, then the queue does not receive a share of leftover bandwidth. If the credit is positive, then the queue does not need to use leftover bandwidth, because it can use its own allocation.

This use of leftover bandwidth is the default. If you do not want a queue to use any leftover bandwidth, you must configure it for strict allocation by including the exact option with the transmit-rate statement. With rate control in place, the specified bandwidth is strictly observed. (On J-series Services Routers, the exact option is useful within a given priority, but not between the priorities. For more information, see Configuring Priority Scheduling.)

On J-series Services Routers, leftover bandwidth is allocated to queues with negative credit in proportion to the configured transmit rate of the queues within a given priority level.

M-series and T-series platforms do not distribute leftover bandwidth in proportion to the configured transmit rate of the queues. Instead, the scheduler distributes the leftover bandwidth equally in round-robin fashion to queues that have negative bandwidth credit. All negative-credit queues can take the leftover bandwidth in equal share. This description suggests a simple round-robin distribution process among the queues with negative credits. In actual operation, a queue might change its bandwidth credit status from positive to negative and from negative to positive instantly while the leftover bandwidth is being distributed. Lower-rate queues tend to be allocated a larger share of leftover bandwidth, because their bandwidth credit is more likely to be negative at any given time, if they are overdriven persistently. Also, if there is a large packet size difference, (for example, queue 0 receives 64-byte packets, whereas queue 1 receives 1500-byte packets), then the actual leftover bandwidth distribution ratio can be skewed substantially, because each round-robin turn allows exactly one packet to be transmitted by a negative-credit queue, regardless of the packet size.

In summary, J-series Services Routers distribute leftover bandwidth in proportion to the configured rates of the negative-credit queues within a given priority level. M-series and T-series platforms distribute leftover bandwidth in equal share for the queues with the same priority and same negative-credit status.

[Contents] [Prev] [Next] [Index] [Report an Error]