3IntroductionThe new enterprise network needs to anticipate customer needs; support seamless collaboration withcustomers, suppliers and employees; and enable exceptional levels of personalized service. Nortel Networksbrings these requirements together and addresses them in an innovative new enterprise strategy centeredaround our vision called “One network. A world of choice.” “One network” because it supports infrastructureconvergence and eliminates boundaries. “A world of choice” because it delivers options on how the enterprisebuilds the optimal network to suit its needs.Nortel Networks enterprise strategy includes Internet protocol (IP) telephony as an enabler of increasedproductivity and increased customer engagement, running on a converged, application-optimized network.IP telephony solutions now scale to 200,000 users to serve telecommuters, remote offices, contact centers, andcampuses. IP telephony has matured to allow full-scale enterprise deployment: centralized or distributedcontrol, enterprise-wide access to applications such as unified messaging, uncompromised voice quality,choice of features and functions, multiple migration paths, and coexistence with legacy systems. In fact, manycustomers are already rolling out IP telephony solutions, aiming to reap the benefits of convergence in the LANand the WAN, and of converged applications. These include the full gamut of Nortel Networks fully-featured,highly reliable and scalable solutions including the i2002/i2004 IP telephones and i2050 soft phones, theBusiness Communications Manager, the CSE 1000, IP-enabled Meridian, CallPilot unified messaging, andSymposium contact center. In doing so, many are asking how to evolve their networks to provide the requiredreliability and performance.This white paper provides technical guidelines on the development of IP networks that consistently, reliably,and securely deliver connectivity, low latency, and throughput for IP telephony. IP telephony is not alone indriving the evolution of enterprise networks from best effort networks designed for data to convergedapplication-optimized networks. Variable and long delays, insufficient throughput, and downtime can result inlost productivity, loss of revenues, and lost customers across the full spectrum of internal and customer-facingapplications. The real-world approach advocated in this paper is based on the overriding principle that anapplication view must be taken in holistically designing the IP telephony system and its underlying network.For IP telephony, this application view starts with the human at the end of the line, while the design of thenetwork must factor in the design of the application.Consistent customerexperience everywhereSecurity for allapplications and servicesBusiness connectivityvia the InternetStorage and networkingat light speedIP telephony succeedstraditional telephonyInternetFigure 1. Nortel NetworksOne network. A world of choice.
4The human factor—requirements and expectationsThe real measure of the performance of IP telephony systems—and of the underlying network—is how well theuser’s requirements and expectations are met, which includes understanding the technical challenges in doing so.The user’s perception, in turn, is affected by sound fidelity, end-to-end delay, and echo. Expectations for reliabilityand security have been set by traditional voice networks (Figure 2). In fact, it has been common practice in theindustry to assess voice quality by asking users what they perceive, quantified as a mean opinion score (MOS) on ascale of 1 to 5, where 5 is very satisfactory. To move away from the subjectivity of MOS, Nortel Networks has beendriving the development of ITU-T G.107, which defines an E-model to objectively measure voice quality. An E-model R value of 70 or greater provides acceptable voice quality.Fidelity (the clarity of the signal) has improved over the decades as the telephone network has moved to digitaloperation. Therefore, the industry talks about toll-quality voice as an objective of IP telephony, referring explicitlyto the user experience over circuit switched networks. Users want this level of fidelity, though they will reluctantlytolerate lower levels if they gain a lot of value (e.g. mobility with cell phones).In IP telephony, voice packets are transmitted over digital transmission facilities with very good error performance;the percentage of voice packets that contain errors (and are therefore discarded) is extremely low. The fidelity of thevoice is dependent on the performance of the coder/decoder (codec) and rate of lost packets. Codecs convert theanalog voice signal to a digitized bit stream at one end of a call and return it to its analog state at the other. Whilebit rates of 64 kbps have been used for years in digital systems, state-of-the-art codecs can deliver near toll-qualityvoice at bit rates as low as 8 kbps (or even lower). The occasional lost packet (e. g., less than one percent) isproblematic for telephony, since this only impacts a short sample of speech; beyond this level, packet loss can bevery disruptive to voice communications. Lost packets arise when noise corrupts the packet or—more likely intoday’s environment—when a switch or router in the path drops packets due to congestion or failure conditions,or when an IP telephone or Media Gateway discards a voice packet that has been delayed beyond some acceptablelimit (as discussed below).Figure 2. Human requirements and expectations for IP telephonyDial tonealways150 msec one-waydelay maxFewimpairmentsCalls areprivateE-ModelR-value of 70
5In audio streaming, one-way delay is generally not an issue. In two-way telephony, delay can impair the quality of theconversation. Delay destroys simultaneity on the call, adversely affecting turn-taking and making it difficult tointerrupt. It can even affect one user’s perception of the politeness, honesty, intelligence, or attentiveness of the other.These impairments become noticeable when the one-way delay rises above 150 ms. This is what is known as the delaybudget. There are various contributors to end-to-end delay, including delays introduced by voice codecs, transmissionand queuing delays, and propagation delays. Once the delay has been introduced, there is no way to remove it, ormitigate its effects.In a lightly loaded network, the one-way delay is the sum of the codec delay; the time required to transmit the voicepacket—including IP, user datagram protocol [UDP], and realtime transport protocol [RTP] headers—on a hop-by-hop basis across the network; plus propagation delay (5 ms per km over fiber or 30 ms across the continental U.S.).That’s the best that can be done and contributes to the end-to-end delay budget. In campus networks, this delay issmall and dominated by codec delay; on the other hand, long-distance calling and the impact of multiple codecs in thepath (for example, transcoding between IP and circuit switching and back to IP) can result in using up to 50 percentor more of the delay budget.Variable packet delay is inherent in all packet networks, due to the receive-queue-and-forward operation that takesplace at every node along the path. IP telephony systems—as well as audio streaming—are designed to compensatefor packet delay variations up to a certain limit. Beyond this limit, received packets are too late to be of value and arediscarded. If this happens occasionally, it probably goes unnoticed. If it happens more often, it can be highlydisruptive.Echo is another related parameter that has an impact on user perceptions of quality. A certain amount of instantaneousecho is required by the human speaker, to avoid the perception of talking into emptiness. Human psychology demandsthat a speaker hear himself speak, so local sidetone is generated by the telephone set. Because of the increased delaycompared to circuit-switched networks, echo control is essential for certain types of IP telephony connections. Thehuman ear and brain can mask undesirable echo; it must be quieter and quieter as the delay increases for the humanperception of performance to remain in the acceptable region.So far, this paper has discussed human perception of voice quality. Users also have expectations that IP telephony isreliable (dial tone is always there and calls are rarely dropped) and secure (existing telephony systems are perceived tomaintain the privacy of communications). The nature of dial-tone is changing with IP telephony since it can begenerated by the device and not necessarily imply service availability; while security in IP telephony is a much broaderissue than privacy.The Appendix provides an overview of the design of the IP telephony systems themselves, which contribute toperformance as perceived by the end user. It does this by identifying a number of best practices of IP telephony systemdesign. IP telephony systems consist of clients, Communications Servers, Media Gateways, and Applications Serversdistributed across an IP network. Signaling and voice payloads are encapsulated into relatively short packets—the latterconsuming 24 or 80 kbps with G.729 and G.711 respectively. The packetization process consumes part of the 150-msone-way delay budget, as do various impairment reduction mechanisms built into IP telephony systems. The reader isencouraged to understand this dimension in designing robust IP telephony solutions.The rest of the white paper addresses the design and engineering of application-optimized IP networks that meetIP telephony user needs, and takes into account the design of IP telephony systems.
6Designing converged networks for IP telephonyTo support IP telephony, a best-effort network—designed for TCP-based data and prone to variable delay and packet lossunder congestion—is being asked to carry applications that require predictable delay and low packet loss (since there is no timeto recover from lost packets). Like some data applications, IP telephony systems have high reliability requirements, and they areassumed to be secure. A systematic approach must be taken in designing such networks starting with the first 100 meters,crossing the campus and the WAN, and ending with a discussion of organizational implications.Nortel Networks has developed a number of guidelines to assist enterprises in designing and building real-world IPnetworks that—together with the design of the IP telephony application itself—consistently, reliably, and securely deliverconnectivity, latency, and throughput requirements for IP telephony and emerging collaborative and engagingapplications.The remainder of this white paper expands on five key areas.The first 100 meters: desktop and wireless LANsA few years ago, desktop networks were built on shared media hubs, using a variety of cabling schemes and with best-effortnetworking. Today, switched Ethernet is the norm, wireless LANs are exploding, and a variety of standards-based Quality ofService (QoS) mechanisms have been built into network products. The following guidelines should be followed in deployingIP telephony in this environment (Figure 3).Structured in-building wiringCategory 5 (or better) structured wiring should be used to the desktop. This will ensure that quality voice can be deliveredover full duplex 10/100-Mbps links. Structured wiring is important in meeting emergency 911 requirements, which requirea correlation between Ethernet port and physical location of the IP telephone.Dedicated switched Ethernet to each telephony desktopOnly switched Ethernet QoS-enabled switching (for example, based on the Nortel Networks BayStack portfolio, including theBusiness Policy Switch) with dedicated ports to each desktop should be used for IP telephony. Shared-media Ethernet hubsmust never be used due to packet collisions that will impact voice quality by dropping voice packets. The Ethernet connectioncould support a soft client in a desktop PC—or separate IP telephone and PC—sharing the port via a three-port QoS-enabledswitch. The wiring closet Ethernet switch should be in a secure location to avoid eavesdropping and other security breaches(which are more difficult with VoIP than with analog phones).IP telephony poweringPower outages pose a serious concern. For certain industries such as health care, even the occasional power outage isunacceptable. In such industries, it is standard practice to provide battery and even generator backup for telephony systems.Powering of IP telephones and the use of uninterrupted power supplies (UPSs) can provide increased reliability for IPtelephony, matching what can be done over private branch exchanges (PBXs). Powering of IP phones can also ease cablingat the desktop. The in-line powering standard is IEEE 802.3af, and is supported by the Nortel Networks BayStack 460PWREthernet switch.
7IP telephony over WLANsWireless LANs operate over a shared radio spectrum, providing mobility for data devices, IP phones, and PC-based soft clients.Running IP telephony on WLANs must address two key requirements—QoS and security over the radio portion. QoS is beingaddressed by IEEE 802.11 for WLANs, which will result in an 802.11e standard. However, Symbol Technologies, Inc.—withwhom Nortel Networks has a strategic alliance focused on IP telephony—has implemented Enhanced Packet Prioritization (EPP)QoS technology in its 11-Mbps AP-1431 Access Point product, which will support 802.11e when standardized. EPP prioritizespacket transmissions from access points to mobile units and is very useful for media content (for example, IP telephony andstreaming video) that can be prioritized over a heavily loaded access point. As with public wireless hot spots, users of QoS-enabled WLANs should expect less than toll-quality voice some of the time, particularly in busy mobile PC-intensiveenvironments. On the other hand, high-quality voice can be expected in controlled environments such as retail. Differentiatedservices (DiffServ) are also supported for end-to-end QoS. Another important consideration with 802.11 WLANs is encryptionand authentication. Native security (for example, Symbol’s MobiusGuard, a comprehensive security suite), wireless applicationprotocol (WAP), or use of IP security measures (IPsec) via IP virtual private network (VPN) soft clients (for example, NortelNetworks Contivity IP-VPN clients) in PCs meet the encryption needs for IP telephony and data alike. For authentication,802.1x and its extensible authentication protocol (EAP) is the recommended approach and is supported by products such asthe Nortel Networks BayStack 470.End-to-end Quality of ServiceMany enterprises have not implemented any form of QoS. Because of this, the traffic may experience differing amounts of packetdelay, loss, or jitter at any given time, which can in turn cause speech breakup, speech clipping, and pops and clicks—or evenworse. Even if bandwidth is over-engineered, growth of traffic, rapid changes of traffic patterns, and network connection failuresmay result in impairments that impact IP telephony (such as packet loss and excessive delays). The following guidelines should befollowed in deploying real-world IP networks that support IP telephony applications across in-building, campus, and broadbandleased lines over the MAN and WAN. Going over public packet data networks and over lower-speed leased lines and Layer 2VPNs requires special attention and will be discussed in the section on QoS and reliability across the public cloud.QoS via 802.1p/QThe IEEE802.1Q standard adds four additional bytes to the standard 802.3 Ethernet frame that provides Ethernet QoS via athree-bit 802.1p field and a virtual LAN (VLAN) ID. Most Ethernet switches—including the Nortel Networks BayStackportfolio—support this standard. Ethernet QoS can be accomplished via the three 802.1p user priority bits, to create eight classesof service for packets traversing Ethernet networks. Ethernet QoS can also be accomplished by prioritizing traffic based on the1 ✒ ✔ ✕ ✖✗ ✘ ✙☛ ✐ ✃1 ✒ ✔ ✕ ✖✗ ✘ ✙☛ ✐ ✃1 2 34 5 67 8 9* 0 #* 0 ##• Category 5 wiring• Dedicated switched Ethernet• QoS: IEEE802.1p/Q• IEEE 802.3af powering• Security: proprietary, WEP, IPsec• QoS: proprietary, IEE802.1p/Q, DiffServFigure 3. First 100 meter connectivityfor IP telephony
VLAN ID only, although with less granularity than using 802.1p. For IP telephony, Nortel Networks recommends an 802.1pbinary value of 110 for both voice bearer and voice signaling. VLANs can be used to separate traffic for ease-of-managementand security purposes, although this is not a requirement. In this case, voice traffic can be placed into one VLAN and non-voice traffic (for example, data or video) into other VLANs. The voice VLAN traffic is prioritized using the 802.1p bits.IP QoS via Differentiated Services (DiffServ)Different types of applications (including IP telephony) have different traffic characteristics and require different types of QoSbehaviors to be applied to them at every router and switch along the path (Figure 4). DiffServ defines a number of differentQoS behaviors and their corresponding QoS mechanisms, called per-hop behaviors (PHBs). These PHBs are identified by anIETF-standardized DiffServ control point (DSCP) carried in each IP packet. Even if there is plenty of unused bandwidthavailable, IP QoS is required, since IP telephony performance may be impacted during times of congestion and traffic peaksand after loss of bandwidth after failures. One of the PHBs defined by DiffServ is the expedited forwarding (EF) DiffServPHB, the behavior of which provides a low-latency, low-loss service that is ideally suited for VoIP. The EF DSCP is representedby the binary value 101110.Figure 4. QoS needs: IP telephony and other applicationsThough more comprehensive schemes can be used to differentiate between various types of data traffic (for example, prioritydata applications) and IP telephony, the simplest approach is to construct network QoS such that there are only two trafficclasses—one for IP telephony and the other for best-effort data traffic. The IP telephony traffic class uses the EF PHB (DSCP101XXX). The best-effort data traffic class uses the default (DF) PHB (DSCP 000000). Separation of voice and voice signalingallows routers and switches along the path to separate these types of traffic onto separate strict priority queues to minimizevoice jitter that would be introduced by the interaction between voice and signaling packets, particularly on slow WAN links.Ethernet switches generally only support IEE802.1p/Q. However, Nortel Networks products such as the Business PolicySwitch and the BayStack 470 also support DiffServ QoS functionality. Passport 8600—as a Layer 2-7 device—supportsDiffServ and much more through its Express Classification (XC) technology.8IP telephony Low High High MediumVideo conferencing High High High MediumSteaming video High Medium Medium Mediumon demandStreaming audio Low Medium Medium MediumeBusiness Medium Medium Low High(Web browsing)E-mail Low Low Low HighFile transfer Medium Low Low HighPerformance dimensionsApplicationBandwidth Sensitivity toDelay LossJitter
9Nortel Networks Service Classes (NNSCs)End-to-end QoS management can be quite complex. Nortel Networks hassimplified QoS by creating standardized, default QoS configurations andbehaviors for its products in the form of end-to-end network service classes.These are called Nortel Networks Service Classes (NNSCs) (Figure 5).NNSCs have been defined based upon the most common types ofapplications. They provide default mapping between DiffServ and differentlink layer QoS technologies that a particular interface uses, such as 802.1pfor an Ethernet interface. NNSCs define default QoS settings per DSCPqueue in which traffic is placed, traffic management parameters, and trafficschedulers. They can also be created on non-Nortel Networks products,through device configuration or QoS policy management systems.The premium NNSC has been defined to be used for IP telephonyapplications such as VoIP. It uses the EF PHB and IEEE802.1p value 6 aspreviously defined. These standards use the strict priority scheduler and apolicer that discards packets that are out-of-profile; that is, that exceed theconfigured bandwidth for the service. Under normal operating conditions,no packets should be dropped. Premium NNSC traffic is also mapped todifferent link layer QoS mechanisms, depending upon the link layer usedfor transport, such as ATM, Frame Relay, point-to-point protocol (PPP),or Ethernet.Figure 5. Nortel Networks Service Class definitionsNortel Networks recommends thatthe four IETF-standardized DiffServPHBs be supported:• Expedited forwarding for IPtelephony applications and services• Assured forwarding (with multilevelrandom early discard [RED]) forvarious types of real-time delay-tolerant and non-real-timemission-critical applications• Default forwarding for best-effortservices• Class selector to migratenon-DiffServ-compliant legacyrouters and switches that onlysupport the IP precedence inthe type of service (TOS) field.A standards-based approach will keepthe IP networking environment open andtherefore most agile to business needs.Network control Critical alarms CriticalRouting, billing, critical OAM NetworkInteractive IP telephony PremiumVideo conferencing, interactive gaming PlatinumResponsive Streaming audio/video GoldeCommerce SilverTimely E-mail, non-critical OAM BronzeFile transfer StandardTrafficcategoryExampleapplicationNortel NetworksService Class
10It is important that all IP telephony packets be queued in a router or switch using a strict priority scheduler, thereby givingtelephony packets priority treatment over all other packets. This is required to minimize voice delay and delay variation (forexample, jitter). Because a strict priority scheduler can starve the servicing of all other traffic queues, a starvation-avoidancemechanism needs to be set to limit the maximum amount of bandwidth that the VoIP traffic can consume. Many products—including the Nortel Networks Passport 8600 routing switch—have this rate-limiting function. In general, weighted schedulerssuch as weighted round robin (WRR) or weighted fair queuing (WFQ) are not recommended. If a router or switch does notsupport a strict priority scheduler (for example, it only supports a weighted scheduler), then the queue weight for VoIP trafficshould be configured to 100 percent. If this cannot be done due to some product limitation, the network provider shouldconsider replacing the product, because it could cause unpredictable voice quality.IP address prioritizationIP telephony traffic can also be prioritized by its IP address. This approach is ideal for devices with statically assigned IPaddresses that rarely, if ever, change. IP PBXs, VoIP gateways, and communications servers are VoIP devices that would havetheir IP addresses statically assigned. Routers and switches can be configured to filter/classify and prioritize all packetsoriginating from these IP addresses.Switch and router performanceEven under heavy load, routers and switches should provide IP telephony traffic with very low latency. In addition, they shouldsupport wire-speed operation (even with short packets) when packet classification (QoS) is activated. Turning on various packetclassification schemes on some software-based routers can have severe impacts on performance, including VoIP packet loss anddelay. This is definitely not the case with the Nortel Networks Passport 8600 routing switch (Figure 6). The Passport ExpressClassification technology provides deep-packet filtering, all in hardware, with no performance degradation even at Gbpsspeeds. Not only is DiffServ supported, but decisions can be made on every packet on whether to allow it, how to queue it,where to forward it, and what changes to make to it.Figure 6. Nortel Networks Passport 8600 Express Classification technologyMarkpacketMark +policePolicepacketService Type 1Service Type 2No Service Type(browsing)Service Type 4Service Type 3FilteringVoice/video/datatrafficService Type 3Markpacket8 Control7 IP Telephony6 eCommerce5 ERP4321 Web surfingOutgoinginterfaceXCC l a s s i f i c a t i o n S w i t c h i n g S e r v i c i n gI N G R E S S E G R E S S
11Expanding QoS beyond IP telephonyThe capabilities described above are designed for IP telephony, but clearly can be expanded across a range of applications,including mission-critical data and real-time collaboration. A comprehensive policy management system is seen as a criticaltool as QoS is expanded beyond IP telephony applications, and is a source of significant operational cost reductions. NortelNetworks Optivity Policy Services manage QoS policies on products such as the Passport 8600, BayRS routers, the BusinessPolicy Switch (BPS), and the Business Communications Manager (BCM).An end-to-end system-level view of reliabilityThe telephony world refers to 99.999 percent base system reliability based on a mean time between failure (MTBF)measured in tens of years and redundant common control (for large systems). But this metric alone doesn’t reflect therealities in real-world IP networks. A few examples can help. An IP network may fail in delivering IP telephony performance:• If it is 100 percent up, but there are non-hardware failure conditions such that a remote site, while physically connected,is logically unreachable (for example, due to routing information protocol [RIP] hop count limits)• If it is 100 percent up, but there is congestion in the network resulting in increased packet loss and excessive delays• If it is 100 percent up, but IP routing convergence after failures takes too longConsequently, for an IP telephony system, the definition of base system reliability is problematic—it is as much a functionof how telephony Communication Server and Media Gateway functions are distributed and designed, as of the underlyingdata-driven infrastructure. Clearly, a comprehensive approach is required to meet the reliability expectations of IP telephonyusers.Traditionally, IP networks achieve reliability through a combination of non-redundant routers running dynamic routingprotocols and applications running error recovery protocols (for example, transmission control protocol [TCP]). However,real-time delay-intolerant applications such as IP telephony don’t run TCP (there’s no time for retransmissions) and,therefore, require rapid recovery from equipment and physical link failures. The following guidelines should be followedin deploying networks which meet IP telephony requirements as they relate to reliability.Backbone node reliability and availabilityBackbone node reliability (driven by MTBF) and availability (driven by mean time to repair [MTTR]) should be headingtowards figures comparable to those for traditional telephony systems, recognizing that networking techniques can be usedto fill the gap. This is achieved by designing switches to deliver the following:• Very high component MTBF• Redundant power, fans, and temperature sensors• Redundant switch fabric and common control with sub-second switchover• Hot swappability of all cards• Automatic short (sub-minute) system boot and restart times• Short (sub-minute) software upgrade service outage timeThe Nortel Networks Passport 8600 is one of the most resilient routing switch products in the industry, and forms thebackbone for both enterprise and carrier networks.
12Rapid detection and recovery below Layer 3IP routing system can take a long period of time to converge after failures. For example, if routing protocols such as openshortest path first (OSPF) are used, the convergence times are proportional to the square of the number of routers in thenetwork, and can last minutes in large networks. Therefore, a sound design principle is to provide resilience at the Layer 1 leveland provide rapid recovery from failures at that level. In this way, link failures can be handled without impacting the Layer 3routing system. Three technologies play key roles in this space:Ethernet link aggregation (cf. IEEE802.2ad) allows multiple 100/1000-Mbps Ethernet links to be configured as a trunk groupbetween wiring closet switches and backbone nodes, and between backbone nodes. Automatic traffic rebalancing takes place ifone of the links fails. Nortel Networks has taken this one step further in its Split Multi Link Trunking (Split MLT) solutionsupported on Passport 8600 and on the BayStack portfolio (Figure 7). Split MLT provides sub-second recovery from linkfailures across trunks homed on two nodes (for extra resilience).Figure 7. Split MLT: A key reliability enablerFor extended campus and data center environments, optical dual ring technologies can provide very high resilience. Theseprovide 50-ms recovery from failures on a SONET and wavelength basis. The Nortel Networks OPTera Metro portfolio isthe foundation for highly reliable storage networking for many of the largest financial institutions.A new option is resilient packet rings (RPR) being standardized by IEEE802.17. RPR is a Layer 2 solution that combinesoptical ring and Layer 2 technology (it is a new medium access control [MAC] layer) to provide 50-ms recovery from failuresby using a counter-rotating ring. Nortel Networks pioneered RPR in its leading OPTera Metro 3500 portfolio.ISTISTC o r e• No single point of failure• Sub-second fail over• Load sharing• Layer 2 recover(no impact on IP)• Interoperable
13Dynamic routing over designed networksSome of the key IP networking standards that enhance fault-tolerant networking include high-performance dynamic routingprotocols (such as OSPF), protocols for route balancing across paths (such as equal cost multi-path [ECMP]), and for LANredundancy (such as virtual router redundancy protocol [VRRP]). These protocols should be carried over networks that aredesigned to put an upper limit on the number of routing points between end users (for example, four, even under single-failureconditions). This puts an upper limit on the delay across the network and speeds up routing convergence times.The capabilities described above are designed to meet the demanding needs of IP telephony, but clearly deliver the benefits ofincreased reliability for all applications running across the network.QoS and reliability across the public cloudMeeting IP telephony QoS, security, and reliability requirements across public packet networks requires special attention.While leased lines are always an option to interconnect sites, virtual private lines using Frame Relay, ATM, and—increasingly—IP-VPNs and Optical Ethernet are attractive alternatives. A high degree of flexibility is required to extend networks reliably andwith the required application-optimized performance across these carrier environments (on possibly a global basis) with theirdiffering interface, signaling, price/performance, and QoS attributes. The following guidelines should be followed in deployingreal-world IP networks that support IP telephony across the cloud.Engineering the bandwidthTypically, LAN bandwidth is inexpensive and is a fixed one-time cost (network interface card [NIC] or switch blade). However,in the MAN or WAN, bandwidth is expensive and results in a monthly recurring cost. QoS allows the enterprise to use expensiveWAN bandwidth most cost-effectively. The bandwidth used for voice calls is dependent on the codecs used and how these areconfigured for different types of calls. How facsimile is handled also needs to be factored in. Traditional voice engineering methodscan be used to determine the number of calls that need to be engineered over the WAN link, factoring in calling communitiesof interest, the number of busy hour call attempts, and the average call holding time. Over under-utilized T3-and-above leasedlines, adding IP telephony traffic uses up available bandwidth. For highly-utilized high-speed links and lower bandwidth (T1or less) connections, the amount of VoIP traffic should be limited to a percentage of the bandwidth of the connection. This isdone to minimize the maximum queuing delay that the VoIP traffic experiences over low-bandwidth connections. For low-bandwidth (less than 1 Mbps) connections, no more than 50 percent of the available bandwidth for voice traffic should beused. For connections more than 1 Mbps, up to 85 percent of the available bandwidth for voice traffic can be used.In packet-based services such as Frame Relay, ATM, and Optical Ethernet, tariffs are based on the access link speed and someform of committed rate and burst size—committed information rate (CIR) in Frame Relay, peak cell rate (PCR) in ATM, andcommitted access rate (CAR) in Optical Ethernet. Adding IP telephony traffic results in the need to subscribe to additionalvirtual circuits (see the following section) and/or higher traffic rate classes.Flexible QoS mapping at the WAN edgeRunning IP telephony over leased lines leaves QoS and traffic management totally under the control of the enterprise, and hasbeen largely discussed previously, with the added comment that special attention must be given to packet fragmentation andreliability as discussed below (Figure 8). Support for flexible QoS mapping when working into carrier packet services is anothermatter, and should be addressed as follows:• Frame Relay standards and products exist for QoS support though service providers have not generally offered QoS-basedservices. DiffServ, in conjunction with Frame Relay traffic management, is used to provide QoS over Frame Relaynetworks. In addition, a separate mesh of virtual circuits (VCs) should be established for IP telephony with appropriate CIR,to minimize interaction between voice and data traffic. The IP telephony VCs should run at a higher priority, if this is
14offered by the service provider. The Nortel Networks Contivity Secure IP Services Gateway portfolio supports securerouting and full QoS support at the edge of Frame Relay networks. A key differentiator of Contivity is that the same devicecan be used over an IP-VPN through a software upgrade.• ATM is designed for multi-service transport, though it is extremely bandwidth-inefficient in supporting IP telephony,unless IP header compression is used. A G.729 IP stream could take up over 80 kbps across ATM. That said, if ATM is tobe used, then IP telephony traffic should be carried over constant bit rate [CBR] or real-time variable bit rate [rt-VBR] VCs(one VC for all voice traffic between a pair of sites). These VCs should be sized appropriately. ATM can support both voiceand data over a single VC, provided that the ATM VC is selected to support the most stringent multiservice application (inthis case, voice).• Optical Ethernet provides native Ethernet connectivity with support for IEEE802.1p/Q. The high-speed, low-latencyattributes of this service make it ideal for MAN/WAN connectivity among metro sites. The CAR may need to be specifiedsuch that it supports the maximum number of simultaneous voice channels plus any data traffic. DiffServ is again used forservice differentiation and IP QoS. The Nortel Networks Optical Ethernet portfolio includes the Passport 8600 andOPTera Metro, and uniquely can be used to build private networks or as the basis of managed services.• Using IP-VPNs over the Internet is very attractive for remote access and for connectivity to remote offices.The implications for IP telephony are described later in this document.Nortel Networks has an extensive set of tools to support its customers in developing real-world networks for IP telephony.Contact your local Nortel Networks representative for details.Figure 8. IP telephony options across public networksCritical CS7Network CS6rt-VBR 7Premium EF-CS5 CBR or(IP telephony) rt-VBR6Platinum AF4x*, CS4 5Gold AF3x*, CS3rt-VBR4Silver AF2x*, CS2 3Bronze AF1x*, CS1nrt-VBR2Standard DE,CS0 UBR 0* x=1, 2, or 3DiffServ CodePoint (DSCP)NNSC ATM servicecategory802.1puser priority
15Reducing delay through packet fragmentationIn mixed voice/data IP networks, packets must be fragmented prior to traversing bandwidth-limited (less than 1 Mbps)connections to minimize voice delay and jitter. There are several different protocols that can be used to fragment packets. ForFrame Relay connections, the provider can use the FRF.12 standard. ATM natively provides fragmentation, since all packets arefragmented into 53-byte ATM cells. However, there are two types of fragmentation that are more universal and not limited toa specific link layer technology such as ATM or Frame Relay—IP and PPP fragmentation. IP fragmentation adjusts the packet(maximum transmission unit [MTU]) size for all packets traversing the router. PPP fragmentation splits large packets intomultiple smaller packets and encapsulates them into PPP frames before queuing and transmission. Recombination is done atthe other end of the link. PPP fragmentation is local-only (as opposed to IP fragmentation, which is source-to-destination),so the two WAN routers initiate and terminate the PPP session. PPP fragmentation allows higher-priority VoIP packets tointerrupt and transmit ahead of the remainder of larger, lower-priority packets that have already been queued. The packetsmay be interleaved so the maximum delay a voice packet will experience is one packet fragment at a time. The fragmentationsize is adjusted to achieve a maximum delay of 20 ms over the different connection speeds. The recommended fragmentationsize is “N” times 128 bytes for a link speed of “N” times 64 kbps (for example, 512 bytes at 256 kbps).Reliability across the WANExtending the reliability of the campus across the WAN can be a major challenge. While IP routing is the last line of defense,lower-layer mechanisms are required to minimize the impacts of failures and meet IP telephony reliability requirements. Withserial links—such as Ethernet on fiber, PPP, Frame Relay, and ATM—various multi-link redundancy options are available(such as Split MLT, PPP multilink, Frame Relay multilink, and ATM inverse multiplexing, respectively). These provide scalablebandwidth and enhanced reliability, though in the case of Frame Relay and ATM, detection of failures and transferring oftraffic to active VCs can take seconds; unlike physical circuits, remote end failures of VCs take time to detect and propagateacross the packet network. With SONET and wavelength rings, and RPR extended to the enterprise site, very high reliabilitycan be delivered with full redundancy and 50-ms recovery times. These styles of optical solutions would be justifiable onlywhen the total needs of a given site are taken into account, including voice, data, video. and storage.Secure IP telephony across the InternetThe Internet is used extensively for employee and partner remote access and connectivity to remote offices, leveraging IP-VPNsfor data applications—for example, using Nortel Networks Contivity client and Secure IP Services Gateway portfolio. It is veryappealing to enhance the productivity of road warriors, telecommuters, and remote office workers by supporting IP telephonyover these IP-VPNs. Converged networking can also have immediate payback by eliminating toll charges. Security concerns ofrunning voice over the Internet can be taken off the table, because all traffic leaving the site across an IP-VPN is authenticatedand encrypted. Reliability for individual users is less of an issue since, in most cases, alternative methods are available (home orhotel phones, and cell phones). For remote offices, redundant access links and dynamic routing over encrypted tunnels (forexample, using Nortel Networks Contivity secure routing technology) can provide a high level of reliability, recognizing thatlocal public switched telephone network (PSTN) interfaces provide an ultimate backup path. QoS is the most thorny issue,since ISPs don’t generally offer QoS, although business-grade IP network offerings do offer service level agreements (SLAs)putting an upper limit on latency (for example, 100 ms) and guaranteeing some level of reliability. Only these kinds of servicesshould be considered for enterprise site-to-site IP telephony, avoiding open-ended performance associated with consumer-oriented networks. In addition, broadband access should be used (such as digital subscriber line [DSL], cable modems, andEthernet) to eliminate access delay bottlenecks. Finally, QoS mechanisms should be used at least for all traffic leaving the site,even if they only apply for the egress queues. Following these guidelines can deliver quality voice a good percentage of the time,and reap economic and productivity benefits. Options to use public voice networks should be retained.
16Organizational implicationsThe greatest technologies will not yield the desired result unless they are engineered and operated appropriately. TraditionalIP networks evolved from PCs to PC LANs to bridged and ultimately switched and routed networks. At the same time,applications running on these networks have evolved from e-mail and file transfers to enterprise resource planning (ERP),supply chain management (SCM), customer relationship management (CRM), and now IP telephony and collaboration.Enterprises recognize that they have had to continuously rethink and evolve their internal procedures and engineeringpractices, in partnerships with vendors, as the importance of the network has grown to become the very life blood of theenterprise.The following guidelines should be followed in deploying real-world IP networks that support IP telephony across the cloud.Network convergence drives organizational convergenceDeploying IP telephony solutions on top of a converged network requires a mixture of skill sets, including a goodunderstanding of what the IP telephony end user wants from feature and performance perspectives, IP telephony applicationengineering, and network engineering, operations and planning. Combining these skills in a single organization can eliminatea number of hurdles in rolling out IP telephony solutions.Designing the network in line with the businessThis white paper has focused on meeting the needs of IP telephony, although it is recognized that the business is puttingtremendous pressure on the network for increased capacity, performance, and reliability across a range of applications. ITplanners must consider networking for IP telephony in the broader context of application-optimized networking across theenterprise. They must establish business-driven reliability objectives, as well as security and QoS policy management directions.On the former point, enterprises need to establish the levels of network-level redundancy that are affordable and justifiable tomeet business needs. For example, they need to continuously evaluate the adequacy of their redundancy plans, across thespectrum from having redundant bandwidth in place to meet business-critical traffic only in case of single failures, to havingredundant bandwidth to carry all traffic even in the case of multiple failures.Operational evolutionEnterprises need to establish operational procedures that recognize the transition from best-effort networking to always-on,application-optimized converged networks. Scheduling maintenance windows and avoiding equipment resets as the first stepfor fault recovery are but two examples of areas that need to be addressed. These need to be consistently applied across theenterprise. Meeting reliability and QoS requirements for voice within the network needs to be complemented bycomprehensive network management tools. These provide configuration management, monitor network operation, andfacilitate rapid fault isolation across multi-vendor network environments. Vendors who have experience in deploying andoperating global enterprise networks can assist in this area.SLA management for converged networksThe increased reliability and performance requirements of converged networks put added pressures for the establishment ofstrong SLAs with service providers. Once established, there is a need to validate that these commitments are being met. Thisrequires a combination of management tools and reporting—generated internally and by the carrier—and a real-time windowthrough service provider customer network management on how the network is performing.
17ConclusionsSuccessful deployment of IP telephony solutions requires an enterprise network that consistently, reliably, and securely deliversconnectivity, latency, and throughput for IP telephony applications. Five key areas need to be addressed:• The last 100 meters through the use of switched Ethernet and wireless LAN• QoS implemented uniformly across the network• End-to-end reliability, recognizing that time is of the essence• Flexibility in interfacing to public networks for high availability and QoS• Review of internal operations, from organization to operational proceduresThe guidelines provided in this white paper allow the enterprise to build real-world networks that support IP telephony andmeet user needs and expectations. These guidelines are real-world, because they take a holistic approach spanning theapplication and network design.Why IP telephony?The transformation of enterprise networks to IP telephony is not just about reducing toll charges or recreating PBXs, butabout re-inventing business communications:• Moving voice to the Web model does for person-to-person communications what it did for information and transactionnetworking. It puts end users in control of their communications, enriches how corporations and governmentscommunicate with customers, and enhances how people collaborate.• IP telephony provides a cost-effective way to provide voice and unified messaging for telecommuters, remote offices,campuses, and contact centers—wherever users might be located.• IP telephony lowers capital and operating costs by converging disparate voice and data onto one network.• IP telephony increases revenues by adding new value to voice applications, such as unified messaging and management,Web-enabled multimedia contact centers, remote PC-based call management, and more.Nortel Networks IP telephony solutionsSuccession enterprise solutions deliver incremental value to any network:• Succession enterprise communication servers are fully distributed IP PBXs supporting a wide spectrum of industry-leadingapplications and features combined with business-grade reliability, investment protection, and global availability. NortelNetworks Succession CSE 1000 is the most feature rich and reliable IP PBX system available for enterprise customers. Itspeer networking functionality exploits the flexibility of IP networks, allowing seamless network integration, simplifiedmanagement, greater flexibility in network deployment, and reduced costs for supporting an increasingly distributed globaluser community.• Succession CSE MX supports new Session Initiation Protocol (SIP)-based multimedia applications (e.g. presencemanagement, collaboration), and bridges the gap between traditional PBXs and next-generation networks. Succession CSEMX brings a new level of feature functionality, scalability, and reliability to converged enterprise networks of all sizes.• Meridian 1 and Norstar are the world leading office communications systems with over 43 million lines installed. IP enablingthe Meridian 1 represents the smoothest evolution path to IP telephony with full investment protection for existing features,telephones, and equipment. Meridian 1 is evolving into an IP-based communications system with all equipment distributableover an IP converged network. Nortel Networks continues to invest in both Meridian and Norstar, and provides investmentprotection as our customers move towards IP telephony.
18• For smaller locations under 150 users, the BCM serves the same multimedia hub function as the CSE 1000 at a lowermaximum station/line capacity and lower price point. With the arrival of BCM 3.0, the capacity of the BCM will beincreased to 200 users, and via the use of the centralized management capabilities, large networks of users can beimplemented using multiple BCMs.• For even smaller locations requiring the services of the enterprise network, the family of Remote Office products isavailable. Remote Office 9150 supports all Nortel Networks Meridian digital stations and services for up to 32 users,and is fully survivable, that is, it can continue to function even if the connection to the core network is lost. Fullysecure configuration can be supported through the Contivity line of products.• CallPilot 2.0 unified messaging is the ideal solution for converging voice mail, e-mail, and fax on a single desktop,vastly simplifying the enterprise messaging experience. Its integrated design strategy allows for full unified messagingdeployment without any impact to existing e-mail servers. As a key application in the Succession enterprise portfolio,CallPilot provides new mobility solutions (e.g. for retrieving voice and e-mail messages with spoken commands overany voice capable device), easy and secure remote worker access, as well as enhanced server capacities and systemfunctionality. Unified messaging with CallPilot significantly improves overall user productivity while lowering theoverall cost of messaging for the enterprise.• Symposium and Periphonics represent Nortel Networks solutions for customer contact management systems.Managing customer contact today plays a vital role in growing revenue while simultaneously lowering cost of sales.Symposium is Nortel Networks contact center system, a fully scalable system for centralized and distributed callcenter arrangements. Symposium goes beyond basic telephony call centers to include fully IP-enabled call centers toprovide features like escorted browsing and “click-to-talk” services from the Web site. Periphonics is Nortel Networkssophisticated line of Interactive Voice Response systems that feature drag and drop management interfaces to instantlybuild customized applications.Why Nortel Networks?For the second year in a row, Gartner Group recognizes Nortel Networks as the clear market leader with the vision andthe ability to execute in the IP telephony market. This is because Nortel Networks is the only provider partner that candeliver solutions combining complete end-to-end networks with advanced voice services and applications. We collaboratewith our customers to unleash their profit potential by providing innovative new technologies as seamless additions toexisting solutions, ensuring investment protection, superior quality and functionality, and graceful deployment options.Nortel Networks cost effectively delivers superior incremental value and choice to our customers, and we can help todrive collaboration, converged services, and improved information workflows across their business, while giving theiremployees the productive freedom to network seamlessly and securely, anytime, anywhere. One size does not fit all—thatis why our converged network solutions feature a range of network elements that have been cost-optimized to deliver thebest possible price/performance regardless of the size requirement of the solution.Forming strategic partnerships with a proven global supplier is an important asset for the enterprise IT manager.Nortel Networks is such a partner—it understands the challenges faced by businesses in developing stronger customerrelationships, and the critical role of IT and the Internet in serving business objectives. The information managementframework is no longer an adjunct support structure; it is the essential foundation for corporate performance. Howinformation is obtained, validated, stored, accessed, and distributed is central to organizational survival and profitability.That’s why we’ve developed our enterprise vision centered around One network. A world of choice.
Nortel Networks:• Has exhibited technology and standards leadership in critical areas, including IP and high performance QoS-enablednetworking and IP telephony and applications.• Has a broad experience base in applications and their networking needs, through its own enterprise network as well as byworking with customers across industries.• Is a proven supplier of highly scalable, reliable multiservice networks supporting voice and data over packet-basedtechnologies, whose products are certified to work with carrier services on a global basis.• Follows strict software and hardware design methodologies for highest quality products, and delivers 7x24 support and afull range of professional services offered on a global basis.Initially, the Nortel Networks strategy means fewer network elements and better capacity management and utilization. Theresults will be seen in lower total cost of ownership and greater flexibility and performance when deploying IP telephony.Ultimately, the rewards will be seen in enterprise evolution to a more profitable and efficient business model that permeatesall aspects of the business and its relations with all stakeholders.Appendix: IP telephony application design best practicesAn IP telephony system is a hardware/software solution that is made up of a set of four logical functions:• IP telephones and PC soft clients• Communications Servers (also called Call Management servers or Gatekeepers)• Media Gateways providing flexible network access (for example, via traditional PBXs, the PSTN, the public wirelessnetwork, and beyond)• Application Servers (for example, unified messaging, conferencing, and SIP-enabled collaborative applications)These functions are distributed across an enterprise IP network, with extended reach and mobility provided over wirelessLANs and the Internet.High-quality multi-mode codecsCodec choice sets the bar for best achievable call quality and drives bandwidth requirements. The two most common codecsused in IP telephony (including Nortel Networks Succession solutions) are G.711 (at 64 kbps) and G.729A (at 8 kbps). Thesebit rates are for the voice payload alone, and exclude Layer 2 (Ethernet or Frame Relay), Layer 3 (IP) and Layer 4 and above(UDP, RTP) overheads. The general tradeoff is that the greater the compression rate, the greater the added delay. Morespecifically, G.729-based systems can use up to 50 percent of the delay budget—five times more than G.711-based systems.Most IP telephony systems support voice activity detection, which suppresses packet transport during silent periods and canresult in 40 percent fewer packets being sent. Most IP telephones support both of these codecs and negotiate which codec isoptimal for a given call. For example, over the WAN, bandwidth is expensive and, depending upon the amount of bandwidthavailable, G.729A may be preferred. On the other hand, interfacing into circuit-switched networks is best handled with G.711to minimize the hit on the delay budget, given the general lack of knowledge of what additional delays will be incurred on theend-to-end connection. (Perhaps the remote end is another IP telephony system requiring the traffic to go through anothercodec.) Codec usage has a significant impact on network engineering of the IP network.19
20Impairment reduction in IP telephony systemsState-of-the-art IP telephony solutions incorporate a number of packet-adaptation and impairment-reduction techniques to improve the quality ofcommunications.Adaptive user/jitter buffersIP telephony systems are designed to compensate for packet delay variations.This is done by time-stamping voice packets, buffering received packets in auser or jitter buffer in the IP telephone or Media Gateway, and playing outpackets in a speech burst as they were received. The length of the buffer definesa play-out window; any packets received outside of this window are discarded,since they arrive too late to be of value. Adaptive operation helps minimize thenumber of late packets that are dropped when the system is congested, andavoids adding unnecessary delay when congestion eases. The buffer is adjustedduring silent periods so the temporal shift in the signal is transparent to users.The added delay is in the 1- to 20-ms range.Packet loss concealmentPackets may be late in arriving and discarded by the receiving device, or mayhave been discarded within the network due to congestion. In any case, themissing information degrades the voice quality. Packet loss concealment (PLC)implemented in the IP telephone attempts to preserve the spectral characteristics of the talker’s voice and maintain a smoothtransition between the estimated signal and the surrounding original samples. Concealment techniques are most effective forup to 60 ms of missing speech. This smoothing comes at a cost of slightly increased delay (10 ms). Nortel Networks productssupport PLC as an integral part of the codec itself and add PLC to improve the performance of G.711 operation.Dynamic echo cancellationEcho control is not required over an end-to-end IP telephony connection. There are various places that echo is generated in anend-to-end connection; for example, at the boundary between the digital network and an analog network. Dynamic operationadapts to network conditions and operates over a wide range of connection types.QoS markingIP telephones, Communications Servers, Media Gateways, and converged applications are all seen as applications running onthe IP network. Because of the time-sensitivity of this traffic, these edge devices pre-mark packets with IEEE 802.1p andDiffServ, to ensure the network elements provide the proper QoS for these packets across the network (the latter beingdiscussed in detail in the next section). Inter-system connectivity typically has involved the use of circuit-switched trunksbetween IP telephony islands, with resulting delay impacts of transcoding from IP to TDM to IP. End-to-end IP telephony andQoS are required to meet voice quality demands in the limited/expensive bandwidth WAN world and to eliminate transcodingdelays. These capabilities are being introduced across the Nortel Networks Succession portfolio, and are differentiated by therich feature set supported. Even with QoS, the right answer under certain circumstances (for example, loss of network capacityunder failure) may be to route voice over the PSTN. For example, the Nortel Networks Remote Office 91XX portfoliomonitors IP network performance and non-disruptively reroutes existing and new voice calls over ISDN B channels.Some of the characteristics of codingschemes commonly chosen for VoIPapplications are as follows:• G.711 is the codec generally used in64-kbps circuit-switched transmission.It is a waveform codec, and, as such, itsoutput is packetized with whateverframe size is required. G.711 can use a10-, 20-, or 30-ms payload size.• G.729/G.729A is the 8-kbps codecstandard. It has comparatively gooddelay characteristics, and has baselinevoice quality falling into theacceptable range. G.729A is a reducedcomplexity version of G.729 withoutquality impacts. G.729 and G.729Aboth use a 10-ms payload size.
21IP telephone acoustic designThe design of IP phones should meet the audio performance standards specified in TIA/EIA-810, which was driven by NortelNetworks. This is the critical last centimeter between the IP telephone and the human ear. TIA-810 is the first all-encompassing standard in the world for audio performance of narrowband digital telephones, including IP telephones. Thisstandard establishes handset, headset, and handsfree telephone audio performance requirements for digital wireline telephones,regardless of protocol or digital format. Loudspeakers will introduce large amounts of echo. Speakerphones are notorious foroften being the source of voice quality degradation, particularly in the generation of echo and distortion when using G.729.Proper acoustic design ensures high-fidelity speakerphone operation under a broad range of conditions. In fact, this is part ofthe Nortel Networks i200X IP telephone portfolio.High-fidelity voice on PCsTo guarantee the correct audio transmit and receive levels, distortion, frequency response, and echo return loss, and to correctlylimit peak acoustic pressure as specified in TIA-810, softphones should be designed as part of a system. High-end sound cardsthat produce great results for music and gaming do not necessarily have the correct characteristics for telephony. When usedwith sound cards and unknown headsets/handsets, softphones will produce unpredictable results. Therefore, any calls madewith these components that terminate on the public network will likely be in violation of FCC and Industry Canada standards.The Nortel Networks Succession i2050 softphone integrates the headset, headset cords, USB adaptor, and softphone audiostack as part of a system that meets the TIA-810 specification. The gains, distortion, echo control, frequency response, andperformance limits of the Nortel Networks USB adaptor match the softphone characteristics. Even the headset cord plays animportant role in echo control.Special handling for fax and DTMF tonesConventional (analog) fax traffic that uses the circuit switched network today is totally intolerant of packet loss and cannot benatively run over IP networks, even with G.711 64-kbps coding. Preserving investment in fax machines is an importantconsideration and is addressed by T.38, an international standard which was driven by Nortel Networks. T.38 describes thetechnical features necessary to transfer facsimile documents in real-time between two standard Group 3 facsimile terminals overthe Internet or other networks using IP protocols. This is done at speeds up to 14.4 kbps. This contrasts with earlier methodsthat used store-and-forward techniques, fitting the e-mail model. The advantages of real-time fax over IP are guaranteeddelivery, easy and familiar operation, and immediate satisfaction. IP telephony solutions, such as the Nortel Networks IP-enabled Meridian 1, have the capability of dynamically detecting fax transmission and invoking T.38 functionality for cost-effective transmission over IP.A dual tone multifrequency (DTMF) tone is what the user hears when depressing a key to access, for example, a voice mailsystem or an online banking system. These tones generally won’t work well across a compression scheme such as G.729. Thesolution is to intercept DTMF tones and convert them into signaling packets.Survivability under network failureCommunications Servers and Media Gateways must be designed to be cost-effectively distributed to meet survivabilityobjectives set by the enterprise. Local and off-net calls can continue to be made even in case of loss of network connectivitybetween sites. This is a key capability of the Nortel Networks Succession portfolio, which provides survivability withoutcompromising feature operation. This ensures that new calls can continue to be established in the event of server failures, andthat existing calls are not impacted. Survivability is a key attribute of the Nortel Networks Succession portfolio, protectingagainst network failures. In the event of server failures (noting that the Nortel Networks Succession CSE 1000 is purpose-builton a real-time UNIX operating system and has an MTBF of 34 years), IP telephones can be configured with primary andsecondary server addresses to support automatic re-registration and seamless switchover in case of failure.