Motorola AT&T NVG589, Руководство администратора

Страница: 128 / 216

Administrator’s Handbook
128
Queue commands
Queue configuration typically requires a classification component to set a QoS marker to a packet and a queueing
component to schedule the marked packets to the link. This is accomplished using filtersets (
“Filterset com-
mands” on page 124 ).
The basic queue 's size and “ length ” are controls for how many packets and total bytes can be enqueued before
it is considered to be full. Once it is full, any attempts to enqueue another packet will result in a “tail-drop.”
Both constraints are simultaneously used, such that it is full when either packet count OR byte count exceeds the
limit. This allows flexibility in obtaining a balance, where a large number of small packets, but only a small number
of large packets can be enqueued.
If there are no tail-drops – that is, the queue is not blocked from sending and doesn't over-fill and dump packets –
then these queue size/bytes parameters do not affect anything. Their only function is to adjust the threshold at
which the queue is considered full, which dictates when tail drops will occur. So if there are no tail-drops, then
increasing the queue length will have no effect. Increasing the queue length has no effect unless there are tail-
drops.
The maximum size/bytes of a queue balances how much burstiness can be buffered versus having a queue that is
simply too long.
Burstiness smoothing
requires queueing up the buffers. For example, if the upstream line rate is 1mbps, but the
traffic source sends 100mbps bursts for 10ms every second (which coincidentally averages 1mbps) then the
router will have to buffer enough (about a full second worth of traffic) so that the burst of traffic doesn't get tail-
dropped when it arrives and is enqueued at the same time in the same burst.
On the other hand, it is undesirable to buffer too much data in the queue(s) since the packets may be stale by the
time they are sent. It may be desirable to drop the traffic sufficiently that there are queuing disciplines such as
Random Early Discard (
red ) that don't drop from the tail of the queue. Instead, red drops packets towards the
front of the queue, so that the congestion is noticed more quickly in order for the sender to scale back bandwidth
usage to avoid drops.
the following types of queue “building blocks” are supported:
◆
basic queue
◆
ingress queue
◆
priority queue
◆
wfq (weighted fair queue)
Basic queues have three different packet dropping options
◆
byte|packet fifo (bpfifo)
◆
random early discard ( red )
◆
stochastic fairness queuing ( sfq )
set queue name queue_name type [ basic | ingress | priority | wfq ]
Sets the type of queue.
set queue name queue_name options [ off | red | sfq ]
Sets the queue packet dropping options.
set queue name queue_name size [ 1... 64 ]
Sets the maximum number of packets that can be enqueued.