User Manual - Configuration Guide (Volume 3)
Versatile Routing Platform Chapter 3
Congestion Management
3-6
III. CQ
Queue 1
Queue 2
Queue 16
...
Class ification
Dequeuing dispatch
Packets leaving the interface
10%
10%
30%
Queue
P acke ts s e n t fr o m this inte rfa ce
Figure QC-3-4 Schematic diagram of custom queuing
As shown in the figure above, CQ classifies the packets into 17 categories
(corresponding to 17 queues in CQ) following the specified policy. According to its
category, the packets queue up and enter the corresponding queue of CQ based on the
first-in first-out policy. Of the 17 queues of CQ, No.0 queue is a system queue, and
No.1 to No.16 are user queues. And the user can configure the proportion of interface
bandwidth occupied for each user queue. When the queues are dispatched, the
packets in the system queue are sent with priority until the queue is empty. Then
according to the bandwidth configured beforehand, a specific amount of packets in
No.1 to No.16 queues are sent out in the polling mode in sequence.
PQ assigns absolute priority to the packets with higher priority level com pared with the
packets with lower priority level. Although this ensures that the key service data c an be
transmitted with priority, the packets with lower priority level will all be congested if the
bandwidth is occupied completely for transmitting massive pack ets with higher priority.
If CQ is adopted, this case can be avoided. There are 17 queues in CQ. The user can
configure the policy of flow classification, and specify the prop ortion of interface
bandwidths occupied by the 16 user queues. Thus, the packets of different services are
allocated with different bandwidths, ensuring that key services can get m ore
bandwidths while preventing from no bandwidth available for ordinar y services.
In the network diagram shown in figure QC-2-1, suppose the server of LAN 1 sends key
service data to the server of LAN 2, and PC of LAN 1 sends ordinary service data to PC
of LAN 2. The serial port connected with WAN is configured with CQ for congestion
management. Key service data flow between the server enters queue A, an d or d inar y
service data flow between PCs enters queue B. The proportion of interface bandwidth
occupied by queue A against queue B is configured to 3:1 (for example, qu eue A in
each dispatching can continuously send 6000 bytes of packets, and queue B in each
dispatching can continuously send 2000 bytes of packets). In this way, CQ treats the
two packet types of different services in different ways. When queue A is dispatched ,
the packets is sent continuously till the number of bytes being sent is no less than 6000
or the queue is empty, then CQ turns to dispatching the next user queue. When queue
B is dispatched, the dispatching will not end till the number of bytes being sent
continuously is no less than 2000 or the queue B is empty. In this way, if congestion
occurs and there are always packets in queue A and B to be sent, from the statistics
result it can be seen that the proportion of the bandwidths acquired by the ke y services
against the bandwidths acquired by the ordinary services is approxim ately 3:1.