Table Of Contents
Conexant Slashes Telecom Costs with Cisco QoS Solutions for Voice, while also Protecting SAP Traffic
Conexant System Inc., a $2 billion Rockwell International spinoff, is the world's largest independent provider of semiconductor system solutions for communications electronics. Conexant is deploying a New World, converged data/voice Frame Relay network made possible by Cisco quality of service (QoS) solutions—including QoS Policy Manager (QPM), Cisco IOS® software, the Catalyst® operating system (OS), Cisco routers, and Catalyst 6000 switches. Conexant experienced significant tangible benefits from implementing QoS. Domestically, for instance, a converged voice and data 256 kilobit per second (Kbps) Frame Relay permanent virtual circuit (PVC) provides Conexant with excellent voice quality and outstanding SAP application performance between the company's Santa Clara field office and Newport Beach headquarters, both in California. Internationally, Conexant is saving $4200 per month in voice toll calls between its Tokyo field office and Newport Beach headquarters, while also prioritizing SAP above background traffic.
The Drive to ConvergenceTwo factors drove Conexant to implement a converged network using Cisco QoS solutions: congestion and cost. As background traffic on its network exploded, Conexant's Frame Relay links became heavily congested. Mission-critical traffic from applications such as SAP Accounts Payable or General Ledger competed with non-critical traffic such as bulk file transfers and casual Internet browsing. During frequent periods of congestion, SAP users experienced long wait times—severely impacting financial updates and reports. Cost also became a factor, as Conexant is motivated to reduce costly voice toll-charges between each of its remote international offices and the headquarters in Newport Beach.
Conexant had initially tried a hardware-based competitive solution, but found that this approach did not scale well across its entire network, was cost-prohibitive, and did not support QoS mechanisms necessary to implement voice across the data network.
Conexant's goals were to reduce phone toll charges, prioritize SAP, and merge voice and data onto one PVC to reduce Frame Relay costs. The ability to merge data and voice would only be possible with robust prioritization and bandwidth management features that would ensure minimal delay and jitter for time-sensitive voice traffic.
"We're in the next phase of networking," says Randy Colvin, Conexant Manager of Network Hardware and Architecture, "where QoS features will transform how we handle all types of traffic over the network. Cisco has been an excellent partner, not only because of its well-known customer support, but also because Cisco provided the features we needed. We need many of the advanced voice QoS features in Cisco IOS Release12.0(7)T and QPM is the only policy manager offering support for features such as IP Real-Time Transport Protocol (RTP) Priority, cRTP, and Frame Relay Fragmentation (FRF.12) (below) that were needed for voice."
Conexant's EnvironmentFor local-area network (LAN) access, Conexant maintains the company's largest LAN of 4000 users at its Newport Beach headquarters. Cisco Catalyst 6500s are the core switches, and the company's engineering LAN includes both Catalyst 6500s in the core and Catalyst 5000s in the wiring closet. A central data center in Newport Beach houses most of the company's servers.
For WAN access, Conexant uses a variety of Cisco routers including 7500, 3600, and 2500 series running a range of Cisco IOS software from 11.2 up to 12.0(7)T. QoS was implemented on Conexant's Cisco multiservice 3640 routers (Figure 1). Domestic circuits include Frame Relay PVCs with speeds from 256 Kbps to T1, with 0 CIR (Committed Information Rate). High-speed circuits exist between the Newport Beach headquarters and San Diego, California (DS3); Newbury Park, California (DS3); and Mexicali, Baja California/Mexico (2 T1s). Internationally, circuits are also Frame Relay and range in speed from 64 Kbps to E1 with non-zero CIR PVCs.
Figure 1 Cisco 3640 Multiservice Routers Classify and Mark Voice and SAP Traffic Bidirectionally on Ingress Ports, then Perform Queuing and Fragmentation on the Egress Ports. The Result is Excellent Voice Quality and High SAP Performance Between Conexant's Newport Beach Headquarters and its Remote Sites.
For major applications, Conexant is running SAP version 3.1H across a logical three-tier, physical two-tier client-server architecture running on a Compaq Alpha back-end server. Conexant's SAP application serves 700 users, 200 of them concurrently. Conexant is also running the Promis application at Newport Beach for manufacturing operations, and is using Lotus Notes for electronic communications across distributed servers at remote locations.
Policy TuningConexant used the consulting services of Netigy (http://www.netigy.com), a premier architect of e-business networks to help the company profile the SAP and voice applications, develop and test the QoS policies, and then implement the policies and perform validation. "Netigy was brought in because, in truth, very few people have the expertise in both network and application consulting like Netigy," says Colvin. "Their assistance saved us a lot of time and, ultimately, money in the implementation process."
"Our implementation methodology has four major phases: planning, design, implementation, and measurement," says Kevin Adams, Principal Consultant, Netigy. "The planning phase culminates in generating a business-requirements document. This is a 10-plus-page document that provides us an implementation road map. The document captures Conexant's business objectives, future needs, WAN, LAN, and application environment, scheduling, etc."
"During the design phase, we simulated Conexant's voice, SAP, and background traffic and developed policies to service the voice traffic while protecting SAP, in competition with the background traffic. The simulation was especially important because we wanted to know exactly how the policy would improve application response time and impact network behavior. The results have been very impressive: We've provided the highest priority for voice, while simultaneously providing dramatic improvement in response time to SAP users."
"The implementation of the actual policy itself contains two components: classification and enforcement."
ClassificationPackets are marked according to the level of priority using the IP type of service (ToS) field. Voice packets are first classified using the RTP port range from 16384 to 32767; and are then marked with IP Precedence five. SAP traffic is classified using the source IP address, as the SAP application is deployed on dedicated servers. SAP is then marked with IP Precedence of four. Traffic classification options also include protocol, host name, source/destination Transmission Control Protocol (TCP)/User Datagram Protocol (UDP) port information, source/destination IP address, or Network-Based Application Recognition (NBAR) for URL and other packet content-based classification.
EnforcementQueuing and other QoS techniques were used for enforcement.
Netigy and Conexant set up three classes of service (three queues per interface):
•Gold—for delay-sensitive, real-time applications such as voice
•Silver—for critical, nondelay-sensitive applications such as SAP
•Bronze—for non-critical applications such as bulk file transfers or casual Internet browsing
For voice traffic, Netigy implemented IP RTP Priority which places voice in a strict priority queue; and this real-time queue will always be serviced before any other traffic so that delay and jitter with voice can be minimized and controlled. Priority queuing had previously been used for voice. Netigy found that, by implementing an enhanced IP RTP Priority policy for voice, Conexant could realize dramatic reductions in packet loss and jitter over priority queuing. (Figures 2 and 3)
Figure 2 Voice Packet Loss was Dramatically Reduced by Implementing an Enhanced IP RTP Priority Policy over the Baseline Priority Queuing Policy.
Figure 3 Jitter was Significantly Reduced with the Enhanced IP RTP Priority Policy; this Enables Higher Voice Quality.
After placing voice in a strict priority queue the next step was to determine an appropriate queuing mechanism for SAP traffic. An analysis of the SAP traffic convinced Netigy that a "softer" queuing technique such as WFQ would be most effective for SAP traffic. The traffic analysis, using packet distribution and transmission frequency for SAP, indicated that SAP generated small low-volume packets, a situation where WFQ works well.
As the SAP packet distribution histogram in Figure 4 shows, the majority of Conexant's SAP packets are small—up to 127 bytes. The low-volume nature of the SAP traffic is illustrated in Figure 5, which shows a small number of packets being transmitted over time and a low consumption of bandwidth. The SAP transmission rate averages only 10 packets/sec and consumes an average of only 22 Kbits/sec. of bandwidth.
Figure 4 SAP Packet Distribution Histogram, Showing a Majority of Small-Sized Packets.
Figure 5 Conexant's SAP Traffic Transmits a Small Number of Packets per Second and Consumes a Small Amount of Bandwidth.
This type of traffic profile lends itself to WFQ, as WFQ gives small low-volume traffic priority over large high-volume traffic. WFQ sets up multiple queues, one for each conversation or flow, and provides higher bandwidth to those queues with a higher IP Precedence setting. The IP Precedence of SAP traffic was set to four so all queues with SAP traffic were given greater bandwidth than background traffic with lower IP Precedence settings.
By adding WFQ, SAP response times improved significantly as shown in Figure 6, from 35 seconds to 25 seconds. The time represents the time required to complete one SAP sales order requiring nine separate transactions. This dramatic improvement was possible even after already placing voice in a strict priority queue.
Figure 6 Adding WFQ Dramatically Improves SAP Response Times.
QoS Policy Manager
Conexant installed the Cisco graphical user interface (GUI)-based QoS Policy Manager (Figure 7), which centralizes and automates policy management. Moreover, QPM dramatically increases the speed and accuracy of defining, validating, configuring, and deploying QoS policies. QPM received Network Computing's Editors' Choice Award: http://www.nwc.com/1024/1024f13.html
Cisco QPM is a GUI- and rules-based policy manager that allows administrators to define, administer, manage, and distribute QoS policies automatically. QPM, running on Microsoft Windows NT, supports a comprehensive list of Cisco routers, Catalyst switches and software releases; and that device information can be downloaded from CiscoWorks2000. QPM is the industry's leading, award-winning, directory-enabled policy manager supporting advanced features such as COPS, RSVP, and Voice. For more information on QPM, see http://www.cisco.com/en/US/products/sw/cscowork/ps2064/prod_literature.html.
Figure 7 Cisco QPM is a GUI- and Rules-Based Policy Manager.
"Before QPM, we had to build and deploy QoS policies one router at a time using the command-line interface (CLI)," says Colvin. "It was time consuming and error prone. With QPM, it's easy to get the routers into the GUI interface, to define the policies for multiple routers, and to then push the polices to all of them at once. It's been a big time saver and has improved policy consistency."
As QoS policies are distributed to network devices, they are converted, by QPM, into specific classification, queuing, limiting, and shaping configuration commands. This reduces the complexity of configuring a mix of QoS features across a myriad of different devices, Cisco IOS versions, and Catalyst OS releases. Figures 8 and 9 illustrate the differences between using QPM versus manually configuring QoS using the CLI.
Figure 8 Shows the Ease of Defining QoS Policies within QPM
Figure 9 Shows the Corresponding CLI Commands that QPM Generates. Without QPM these Commands Would Need to be Manually Generated and Entered.
Results"We have some very tangible results from implementing Cisco QoS features," says Colvin. "For example, we are now running data and voice over one PVC between Santa Clara and Newport Beach. What's more, we are getting excellent voice quality over the link and superior performance of the SAP applications. Users don't even know that their call is being routed over the data network, and that's impressive.
"While we are seeing obvious savings from the elimination of the PVC domestically, we're seeing the most dramatic savings internationally across our Tokyo/Newport Beach connection. Currently, we are using two PVCs. But, thanks to QPM, we've been able to move our voice traffic that was previously handled over the PSTN onto the existing PVCs. This is saving Conexant $4200 a month in toll calls, for just one link. In the near future, we expect additional savings of approximately $1000 a month when we eliminate the second PVC. We've been pretty happy with our QoS and QPM implementation."