Table of Contents
Advertisement
Layer 2 QoS
On Avaya and Cisco switches, IP Telephony traffic can be assigned to higher priority queues.
The number and the sizes of queues and how the queues function are device dependent, and
beyond the scope of this document.
However, in general, a fixed number of queues exist, and the queues are usually not
configurable. If the queues are configurable, it is typically not recommended. Older or lower end
switches commonly have only two queues or none at all. Newer or higher-end switches
commonly have four or eight queues, with eight being the maximum because there are only
eight Layer 2 priority levels. When configured to do so, the Ethernet switch can identify the
high-priority traffic by the 802.1p/Q tag, and assign that traffic to a high-priority queue. On some
switches, a specific port can be designated as a high-priority port, which causes all traffic that
originates from that port to be assigned to a high-priority queue.
Layer 3 QoS
It is usually more complicated to implement QoS on a router than on an Ethernet switch. Unlike
Ethernet switches, routers do not just have a fixed number of queues. Instead, routers have
various queuing mechanisms. For example, Cisco routers have standard first-in first-out
queuing (FIFO), weighted fair queuing (WFQ), custom queuing (CQ), priority queuing (PQ), and
low-latency queuing (LLQ). LLQ is a combination of priority queuing and class-based weighted
fair queuing (CBWFQ), and it is Cisco's recommended queuing mechanism for real-time
applications such as IP Telephony. Each queuing mechanism behaves differently, is configured
differently, and has its own set of queues.
First, the desired traffic must be identified using DSCP, IP address, TCP/UDP port, or protocol.
Then the traffic must be assigned to a queue in one of the queuing mechanisms. Then the
queuing mechanism must be applied to an interface.
The interface itself might also require additional modifications, independent of the queuing
mechanism, to make QoS work properly. For example, Cisco requires traffic shaping on Frame
Relay and ATM links to help ensure that voice traffic is allotted the committed or guaranteed
bandwidth. Cisco also recommends link fragmentation and interleaving (LFI) on WAN links
below 768 kbps, to reduce serialization delay. Serialization delay is the delay that is incurred in
encapsulating a packet and transmitting it out the serial interface. It increases with packet size,
but decreases with WAN link size. The concern is that large low-priority packets induce
additional delay and jitter, even with QoS enabled. This is overcome by fragmenting the large
low-priority packets and interleaving them with the small high-priority packets, thus reducing the
wait time for the high-priority packets.
Table 72: Serialization delay matrix
Layer 2 QoS
on page 314 lists
Issue 3.4.1 June 2005
313
Advertisement
Table of Contents
