QoS Support on the Catalyst 2900XL and 3500XL 79
Congestion Management
The 2900XL and 3500XL switches use a shared memory buffer system because each
individual port does not have its own output queue. This shared memory buffer is divided
into two global transmit queues. Each ingress packet is placed into one of two global
transmit queues based on CoS value for tagged frames and CoS classification for untagged
frames. One of the transmit queues is designated for packets with a CoS value of 0 to 3, and
the other transmit queue is reserved for packets with a CoS value of 4 to 7. The queues use
a 100-percent threshold value. These queues are not configurable for different CoS values
or thresholds. This queue scheme creates a logical high-priority and low-priority queuing
mechanism. Priority scheduling is applied such that the high-priority queue is consistently
serviced before the low-priority queue. The use of two global transmit queues based on CoS
value is default behavior and cannot be altered. As a result, no global configuration is
required to enable QoS output scheduling.
NOTE Untagged packets that are classified with a CoS value transmitted on trunk ports are
appropriately tagged with an 802.1q header with the respective CoS. For packets
transmitted on nontrunk ports, the untagged classification only determines which queue the
frame is placed in for egress transmission.
Case Study: Classification and Output Scheduling on Cisco
Catalyst 3500XL Switches
To demonstrate classification and output scheduling on the Catalyst 3500XL series, a
Catalyst 3524-PWR-XL was set up with two Cisco 7960 IP Phones, a Call Manager, and a
traffic generator connected to three Fast Ethernet ports and a Gigabit Ethernet port, respec-
tively. Figure 3-4 shows this topology. Two trials were conducted taking voice quality
statistical measurements from each IP Phone based on a 1-minute, G7.11 voice call
between IP Phone 1 and 2. To create traffic congestion, the traffic generator attached to
Gigabit Ethernet port was sending multicast at line rate with a CoS value of 0. The multicast
traffic was flooded to all ports, including the Fast Ethernet IP Phones, causing output
congestion.
80 Chapter 3: Overview of QOS Support on Catalyst Platforms
Figure 3-4 Catalyst 3500XL Case Study Network Diagram
The Catalyst 3524XL switch was running software version 12.0(5)WC5 for the trial. The
configuration only included voice VLANs on the Cisco IP Phone ports. The remaining port
configuration of the switch was default. Example 3-3 shows the relevant configuration.
Example 3-3 Catalyst 3548XL Switch Port Configuration for Case Study
Switch#show running-config
Building configuration...
Current configuration:
(text deleted)
interface FastEthernet0/1
switchport access vlan 70
!
interface FastEthernet0/2
switchport trunk encapsulation dot1q
switchport trunk native vlan 2
switchport mode trunk
switchport voice vlan 70
spanning-tree portfast
!
interface FastEthernet0/3
switchport trunk encapsulation dot1q
switchport trunk native vlan 2
switchport mode trunk
switchport voice vlan 70
spanning-tree portfast
!
(text deleted)
interface GigabitEthernet0/1
switchport access vlan 70
(text deleted)
end
Catalyst 3524-PWR-XL
Call
Manager
Traffic Generator
IP Phone 2IP Phone 1
Voice Stream
(1 to 2)
Voice Stream
(2 to 1)
Multicast
Stream
Fa0/1
Fa0/2
Gi0/1
Fa0/3
QoS Support on the Catalyst 2900XL and 3500XL 81
The variant in the two trials was the CoS value placed on the telephony frames between the
IP Phones. With a CoS value of 0, the telephony stream was treated with a low priority (the
same priority as the multicast traffic). With a CoS value of 5, the telephony stream was
treated with a high priority. Table 3-7 summarizes the number of frames transmitted and
lost as well as jitter from each trial.
As indicated in Table 3-7, the Catalyst 3524XL did not drop a single frame due to output
congestion on the IP Phone ports for packets with a CoS value of 5. Similar results are
achievable with multiple Cisco IP Phones in a campus network using the Catalyst 2900XL
and 3500XL. The jitter did not vary between the trials because all Catalyst switches drop
frames under congestion and only buffer a few frames. The maximum recorded jitter is
around 50 ms, which is above the recommended 30 ms for Voice over IP (VoIP). Only the
first few frames of the IP flow recorded jitter near 50 ms.
Summary
The Catalyst 2900XL and 3500XL suit basic QoS needs for an access layer switch. If
additional features such as policing and classification based on DSCP are required, network
designers should consider the Catalyst 2950 and 3550 switches for use as an access layer
switch. You can summarize QoS feature support on the Catalyst 2900XL and 3500XL
switches as follows:
• No support for input scheduling.
• Classification based on CoS only; no support for classification based on IP precedence
or DSCP.
• Two global queues for high-priority and low-priority traffic.
• No configurable CoS mapping to queues or queue threshold.
• Ports are trusted by default.
• Untagged frames are mapped to high-priority or low-priority queues based on
configured classification CoS value.
• The Catalyst 3548XL and 3524-PWR-XL support reclassification of tagged frames.
Table 3-7 QoS Trial Results on Catalyst 3524-PWR-XL
Trial
Total Frames
Transmitted (Phone 1/2)
No. of Receive Lost
Frames (Phone 1/2)
Maximum Recorded
Jitter (Phone1/2)
CoS = 0 on
voice stream
3100/3110 1551/1549 51/49 ms
CoS = 5 on
voice stream
3104/3106 0/0 51/49 ms
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.