Alcatel-Lucent 7950 Quality Of Service Manual page 302

Service Queue QoS Policy Commands
bcg
Syntax bcg burst-control-group-name [member-queues [at-risk-only]] [exp-util-bw megabits-per-
second]
Context show>qos
Description
The show qos bcg command outputs the current and historical visitation time associated with the specified
BCG name.
A Burst Control Group (BCG) represents a list of queues that share the same non-scheduling PIR and CIR
bucket target update interval. When a queue's scheduled rate bursts above its PIR bucket depth, the queue is
removed from its scheduling context. The system uses a BCG in order to visit the queues PIR bucket to
periodically drain an appropriate amount from the bucket. When the bucket has been drained below the PIR
bucket threshold, the queue is allowed back onto its scheduling context. The amount decremented from the
bucket is a function of the amount of time that has elapsed since the last bucket update and the queue's
shaping rate (PIR). If the queue's shaping rate is configured as 1Mbps and 1ms has elapsed since the last
bucket update, the system will decrement the PIR bucket by 125 bytes. One caveat is that the bucket cannot
be decremented past a depth of 0. This fact drives how the system chooses which BCG is used to manage the
queue bucket update interval.
If a queue's shaping rate is 1Mbps and the threshold (burst limit) is set to 10Kbytes, the maximum amount
of time that can expire before the queue is updated without resulting in a negative bucket depth is 81.92ms.
This can be calculated by taking the number of bits represented by the bucket depth (10Kbytes = 10 * 1,024
* 8 = 81,920 bits) and dividing it by the rate (81,920 bits / 1,000,000 bits per second = 81.92ms). The queue
will not be removed from the scheduler until the PIR bucket depth has equaled or exceeded the configured
burst threshold, so the bucket will be at least 10Kbytes deep. If the system visits the queue PIR bucket
within 81.92ms, the resulting decrement operation will leave the bucket. If the system takes longer than
81.92ms, the decrement result will be greater than 10Kbytes and part of the decrement result will be lost.
The net result is from less than timely updates is that the queue will not be returned to the scheduler context
fast enough and some shaping bandwidth for the queue will be lost (underrun the shaping rate).
Each Q2 based forwarding plane maintains 7 Burst Control Groups, each targeting a certain queue bucket
visitation time. A 40ms, 20ms, 10ms, 5ms, 1ms, 500us and 100us BCG is supported. By default, queues are
placed on a BCG based on shaping rate and the queue's burst limit (PIR threshold depth) is set based on the
BCG visitation time and the queue's specified shaping rate. When all shaping queues on a Q2 are left in a
default burst tolerance management state, the system has sufficient BCG visitation resources to ensure that
all queues do not experience inaccurate bucket decrement conditions.
When explicit burst-limit threshold values are defined for a shaping queue, the system picks an appropriate
BCG based on the queue's configured shaping rate and the explicit threshold to find a BCG with the best
target visitation time that results in worst case decrement values that are less than the configured threshold.
However, when a queue is placed on a 'faster' BCG, more visitation resources are consumed and it is
possible that the system will not meet a queue's decrement constraints.
The show qos bcg command allows visibility into a BCG's historic and current visitation time. The system
samples the amount of time it takes each list to visit each of its associated queues once each second and
stores the last 10 samples. It also keeps the longest visitation time seen since the last time the BCG statistics
were cleared, the longest visitation time for the current queue-to-BCG lists associations, calculated longest
visitation time based on maximum scheduling bandwidth and lastly the longest visitation time for an
optionally defined scheduling rate.
With each sample, the system indirectly calculates the amount of scheduling bandwidth based on how much
Page 302 7950 XRS Quality of Service Guide