Information Technology Reference
In-Depth Information
It is also possible to specify a minimum amount of bandwidth in kilobits per second for a particular class, as well as the percentage of
remaining available bandwidth using the
bandwidth
kbps
and
bandwidth remaining percent
percentage
commands, respectively.
As mentioned, although CB-WFQ can offer minimum bandwidth guarantees, it cannot offer any latency guarantees; for that, LLQ is
needed. A sample configuration for LLQ is shown in the following example:
! <class maps omitted for brevity>
!
policy-map
llq-policy
class
voice
priority percent
15
class
video
priority percent
10
class class-default
fair-queue
!
interface serial 2/0
ip address
192.168.2.1 255.255.255.0
service-policy output
llq-policy
In the example, the policy (
llq-policy
) specifies a priority queue for voice with a guaranteed allowed bandwidth of 15 percent of link
bandwidth (
priority percent
percentage
). A priority queue is also configured for video and assigned 10 percent of link bandwidth.
Next, a fair queuing strategy is specified for all other traffic; this traffic is serviced only when the priority queues have first been
serviced up to their assigned bandwidth. Finally, the policy map is attached to the interface.
Note that, in LLQ, there is an implicit ability to accommodate bursts of up to 200 ms of traffic. LLQ also includes an implicit policer
that ensures that excess traffic assigned to priority queues is dropped (thus ensuring that traffic assigned to other queues can be
serviced and that priority queues do not monopolize bandwidth).
Although queuing and scheduling mechanisms allow traffic to be buffered and transmitted, queues can sometimes overfill and traffic
can be tail dropped from the back of queues or be selectively dropped before the queues are completely full.
Selective packet dropping allows congestion avoidance and can often allow better performance than tail dropping for data applications
because TCP windowing mechanisms ensure that at any one time only those TCP senders whose traffic has been dropped will reduce
their throughput. Congestion avoidance using selective packet dropping can be achieved on Cisco routers using a mechanism called
weighted random early detection (WRED).
