Performance Measurements of Advanced Queuing Techniques in the Cisco IOS

Multimedia Applications in Today's Internetworks

See Also:

Delivering Consistent Quality of Service

Most of today's data networks carry traffic that is "bursty" in nature. Bursty traffic can affect the perceived quality of service of all traffic on the network by introducing inconsistent latency, also known as "jitter." This can cause problems for some emerging network applications, such as desktop videoconferencing and high-speed video servers, which require a high degree of predictability from a network. The Cisco Internetwork Operating System(tm) (Cisco IOS) addresses this challenge with advanced queuing techniques that guarantee consistent high-quality service without introducing jitter -- even in the presence of bursty traffic.

This paper will describe these queuing techniques and evaluate test results of their performance characteristics. All tests were performed using traffic generated by real multimedia networking applications.

Traffic Prioritization

The need to prioritize packets arises from the diverse mixture of protocols and their associated behaviors found in today's data networks. Different types of traffic that share a common data path through the network can impact each other. Depending on the application and overall bandwidth, users may or may not perceive this impact negatively. For instance, interactive, transaction-based applications may require a higher priority than say, a file transfer. Desktop videoconferencing requires a minimum amount of bandwidth to perform acceptably. If your network is designed so that multiple protocols share a single data path between routers, prioritization may be a requirement.

Effective Use of Traffic Prioritization

Prioritization is most effective on WAN links where the combination of bursty traffic and relatively lower data rates can cause temporary congestion. Depending on the average packet size, prioritization is most effective when applied to links at T1/E1 bandwidth speeds or lower. This is because of the computational overhead required to examine each data packet and then make a queuing decision based on its priority. The Cisco IOS(tm) provides two advanced queuing techniques that can be used to prioritize traffic.

Custom Output Queuing

With custom output queuing, a "weighted fair" queuing strategy is implemented for processing interface output queues. LAN administrators can control the percentage of an interface's available bandwidth used by different types of traffic. If no traffic exists for one of the specified queues, that queue's bandwidth is made available for all other traffic.

Priority Output Queuing

Priority output queuing provides a mechanism to use strict priority in selecting which packets to send first on an interface. This technique is useful in environments where traffic has a hierarchy of importance, and more important traffic should not be delayed by less important traffic.

Priority queuing is commonly used to improve the network performance by prioritizing interactive traffic over batch traffic.

Performance Measurements

The following test results show the impact of using these queuing mechanisms on leased-line, wide-area links. In all test scenarios, full quality of service was maintained for the video and audio streams passing through the network.

Testbed Configuration 1

In the first test scenario, two Ethernet-based Cisco 2500 routers were connected by a T1 (1.544-Mb) WAN link. Each of the routers' Ethernets had two clients attached, one PC and one Mac. The clients were running two desktop videoconferencing applications with audio/video streams destined to their respective PC or Mac peers on the other end of thelink (see Figure 1). The routers were configured with custom queuing allocating 10 percent of the T1 link to each of the two videoconferencing applications. The remaining 80 percent was allocated for all other traffic.

Bro_WP_advque_fig1

Figure 1: Testbed Configuration 1

A network monitoring device was placed in line with the T1 link to measure link utilization. The two bidirectional UDP/IP audio/video streams generated by the videoconferencing applications loaded the network link to 12 percent utilization. The first point that is interesting to note is that existing desktop videoconferencing applications can be easily accommodated on a T1 link. Requiring only 12 percent link utilization for two point-to-point conferences, a LAN administrator could allocate one-third of available T1 bandwidth to support six such bidirectional sessions simultaneously.

Traffic (starting at 64-byte packet sizes) was then injected to simulate a saturated WAN link. Packet sizes were incrementally increased, with measurements being taken at each increment. Figure 2 shows link utilization at different packet sizes.

Bro_WP_advque_fig2

Figure 2: Testbed 1 Results

While injecting 64-byte packets, less than 50 percent link utilization was achieved because of the processor overhead associated with examining many packets per second. Utilization increased to link saturation levels when packet sizes increased to 256 bytes. These results show that as long as average packets are larger than 200 bytes, no significant performance degradation is experienced. The typical average packet size in a network running mixed protocols/applications will be around 500 bytes.

The complete configuration listing for this testbed can be found at the end of this document. For more information on router configuration, refer to Cisco's Router Products Command Reference.

Testbed Configuration 2

Test scenario number two was run to evaluate the effectiveness of advanced queuing techniques as the number of network connections is increased. In this scenario, an additional T1 link was added between the two Cisco 2500 routers (see Figure 3).

Bro_WP_advque_fig3

Figure 3: Testbed Configuration 2

With a second link in place, traffic was equally load-balanced between the links, a feature that is automatic with the Cisco IOS. Again, a measurement was taken to evaluate link utilization for the UDP/IP traffic streams generated by the combined videoconferencing applications. This measurement decreased to 6 percent link utilization per link, showing that the traffic was being load balanced between the links. Again, traffic (starting at 64- byte packet sizes) was injected to saturate both WAN links. Packet sizes were incrementally increased, with measurements taken at each increment. Figure 4 shows link utilization at different packet sizes.

Bro_WP_advque_fig4

Figure 4: Testbed 2 Results

With the additional WAN link in place, two times the amount of injected background traffic was required to fill the WAN links. This additional traffic imposed additional computational overhead on the router, causing the overall throughput to drop. Link saturation was not achieved until packet sizes exceeded 512 bytes. Again, traffic prioritization becomes more effective as the average packet size increases.

The complete configuration listing for this testbed can be found at the end of this document. For more information on router configuration, refer to Cisco's Router Products Command Reference.

Testbed Configuration 3

In test scenario number three, a Cisco 4500 was connected via two T1 WAN links to two Cisco 2500s. A PC client was installed on the first of the 4500's two Ethernets, with a Mac client on the second. The PC and Mac peers were then placed on each of the two Cisco 2500s' Ethernets (see Figure 5).

Bro_WP_advque_fig5

Figure 5: Testbed Configuration 3

A link utilization measurement showed approximately 6 percent of available bandwidth used on each of the two T1 links while videoconferencing sessions were active between the Mac and PC peers. Again, traffic (starting at 64-byte packet sizes) was injected to simulate saturated WAN links. Packet sizes were incrementally increased, with measurements being taken at each increment. Figure 6 shows link utilization at different packet sizes.

Bro_WP_advque_fig6

Figure 6: Testbed 3 Results

The addition of a Cisco 4500 and a second Cisco 2500 causes the overall throughput to increase compared with testbed two. Link saturation was achieved at 512-byte packet sizes.

The complete configuration listing for this testbed can be found at the end of this document. For more information on router configuration, refer to Cisco's Router Products Command Reference.

Testbed 1 Router Configuration

Router 1

version 10.0 ! hostname router1 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.1.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.1 255.255.255.0 custom-queue-list 1 ! interface Serial1 shutdown ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Router 2

version 10.0 ! hostname router2 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.3.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.2 255.255.255.0 custom-queue-list 1 ! interface Serial1 shutdown ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Testbed 2 Router Configuration

Router 1

version 10.0 ! hostname router1 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.1.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.1 255.255.255.0 custom-queue-list 1 ! interface Serial1 ip address 144.254.3.1 255.255.255.0 custom-queue-list 1 ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Router 2

version 10.0 ! hostname router2 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.4.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.2 255.255.255.0 custom-queue-list 1 ! interface Serial1 ip address 144.254.3.2 255.255.255.0 custom-queue-list 1 ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Testbed 3 Router Configuration

Router 1 Cisco 4500

version 10.0 ! hostname router1 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.1.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.1 255.255.255.0 custom-queue-list 1 ! interface Serial1 ip address 144.254.3.1 255.255.255.0 custom-queue-list 1 ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Router 2 Cisco 2500

version 10.0 ! hostname router2 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.4.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.2.2 255.255.255.0 custom-queue-list 1 ! interface Serial1 shutdown ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Cisco 2500 Router 2

version 10.0 ! hostname router2 ! enable password cisco ! ! interface Ethernet0 ip address 144.254.5.1 255.255.255.0 no mop enabled ! interface Serial0 ip address 144.254.3.2 255.255.255.0 custom-queue-list 1 ! interface Serial1 shutdown ! router igrp 109 network 144.254.0.0 ! queue-list 1 default 3 queue-list 1 protocol ip 1 udp 5715 queue-list 1 protocol ip 2 udp 7648 queue-list 1 queue 1 byte-count 19300 queue-list 1 queue 2 byte-count 19300 ! ! line con 0 line aux 0 line vty 0 4 password cisco login ! end

Posted: Tue Jul 20 16:35:39 PDT 1999

Performance Measurements of Advanced Queuing Techniques in the Cisco IOS

Delivering Consistent Quality of Service

Traffic Prioritization

Effective Use of Traffic Prioritization

Performance Measurements

Testbed Configuration 1

Testbed Configuration 2

Testbed Configuration 3

Testbed 1 Router Configuration

Testbed 2 Router Configuration

Testbed 3 Router Configuration

Copyright 1996 © Cisco Systems Inc.