Queuing Mechanisms

Frame and packet marking
Queuing Mechanisms
A number of different queuing mechanisms exist on Cisco layer 2 and layer 3 switches. The reason
for this is that across the globe, different network managers require different prioritization
for different networks running a wide variety of legacy, common, and emerging applications.
No single queuing mechanism could support these diverse needs, so several mechanisms exist.
It is up to the intelligent network administrator to apply the method available for their network
that best suits their needs. Here is a short list of the most prevalent methods available across the
spectrum of Cisco layer 2 and layer 3 switches:
First in, first out queuing First in, first out (FIFO) queuing transmits frames/packets according
to the timed arrival of the first bits in the frame/packet at the input interface. This is often
the default method.
Weighed fair queuing Weighed fair queuing (WFQ) places data into different queues
according to a conversation index associated with each packet. The conversation index is
a term applied to different applications, whose packets are then marked with a number
inside the switch or router. The selection of the data type and queue is internal and proprietary,
but results in low-volume interactive traffic (voice) being granted higher priority than
high-volume non-interactive traffic (FTP).
Custom queuing Custom queuing allows administrators to create up to 16 queues, each with
configurable sizes and forwarding thresholds. Data is placed in queues according to access lists,
and queues are emptied on a round-robin basis.
Priority queuing Priority queuing allows the administrator to create a number of queues
and configure the size of each. Data is placed into queues according to access lists, and
queues are emptied on a strict priority basis. Packets in the highest priority queue are always
transmitted first, and packets in lower priority queues are not transmitted until the queues
higher up are emptied.
Layer 2
header IP Header Data
Encapsulated Packet
Preamble
Start frame
delimiter DA SA Tag PT Data FCS
Layer 2 802.1q/p Frame
Version
length
ToS
(1 byte) Len ID Offset TTL Proto FCS IP-SA IP-DA Data
Layer 3 IPv4 Packet
3 bits used for CoS (user priority)
DSCP
Weighted round-robin queuing Weighted round-robin (WRR) queuing is a simplified version
of custom queuing. A fixed number of queues are serviced in round-robin fashion, each being
configurable only as to the size of the queue.
Multistage queuing Multistage queuing can be implemented on some platforms, and involves
the creation of multiple queuing processes in a dependency fashion. For example, a mixture of
priority and WFQ could be used.
Figure 20.9 shows how packets arriving at three interfaces at the same time need to be sorted
into an output queue to be transmitted serially. Of course, it is in the output queue that we can
influence packet delay by arranging how the queue works.