April 1998

BY TONY RYBCZYNSKI

In the early 90s, the only viable way of supporting enterprise voice traffic was by using circuit-based techniques (i.e., 64 Kbps circuit switching and time division multiplexing). Packet switching (SNA, X.25, IP, and frame relay) was a data-only solution. Most corporations had disparate voice and data networks with no sharing of bandwidth between these two traffic types, either by running over different transport networks or over dedicated pipes on a common TDM multiplexer backbone.

A revolution is in progress that is dramatically changing the way voice traffic is handled. A new wave of enterprise network consolidation using packet technologies has emerged driven by opportunities for improved price and performance.

Voice traffic includes real-time speech (telephony or voice communications integrated with video conferencing), non-real-time voice response (e.g., voice mail access), fax, modem, and digital data. The common attribute is that it is carried over a circuit-switched network such as a private PBX network or the Public Switched Telephone Network (PSTN). The interest in packet voice is spurred by a number of complementary developments, namely:

• The emergence of multimedia PCs and applications.
• The explosion in the Internet and its compelling economics for all sorts of connectivity (including voice).
• The evolution of packet networking technologies to support multimedia: ATM by design, frame relay through the addition of multimedia classes and switched virtual circuits (SVCs), and IP networking through emerging mechanisms such as Resource ReSerVation Protocol (RSVP).
• Variable bit rate voice coding and compression technologies that can provide toll quality voice at considerably less than 64 Kbps and that are suited for packet networking.

VARIABLE BIT RATE VOICE CODING
There are a number of advantages to using variable bit rate voice coding and compression technologies for voice. These recognize the inherently bursty nature of voice communication, which has been characterized generally as being 50–60 percent silence periods, with talking (non-silence) intervals being split between essential components (critical for coherent communications) and repetitive patterns (being redundant). Taking advantage of these characteristics can result in increased networking efficiency, particularly over packet networks, and can improve the performance of data traffic by making all the bandwidth available to data when no one is talking.

Variable bit rate coding schemes can also provide the network operator the opportunity to free up bandwidth under network congestion conditions. For example, with the onset of congestion, increased levels of voice compression could be dynamically invoked, thus freeing up bandwidth and potentially alleviating the congestion condition, while diminishing the quality of the voice during these periods.

PACKET VOICE NETWORKING
Enterprise users may be understandably confused by this explosion in options for handling voice traffic on data networks; after all, "if it isn’t broken, don’t fix it." At the same time, they are intrigued by opportunities to significantly decrease the cost per minute of handling voice traffic (which still accounts for up to 50 percent of their communication costs). And, they are increasingly excited about transforming their often disparate voice, data, and video networks into a single integrated multimedia network.

Are all of these packet voice networking alternatives equally applicable to all situations? And if not, when should each of these be considered? Three distinct packet voice networking environments exist:

• Interconnection of circuit-switched telephony systems (e.g., PBXs) over packet networks.
• Direct end device attachment to packet networks, the end device typically being a PC.
• Internet telephony.

Only the first case (interconnection of PBXs over packet networks) is a mainstream scenario of most immediate importance to enterprise users.

INTERCONNECTING PBXS OVER PACKET NETWORKS
PBXs provide a circuit-switched communications environment among users’ end devices (e.g., telephones) connected via analog or digital (e.g., ISDN) lines. Two distinct approaches for PBX wide-area networking have been widely deployed: specifically, private voice networks and virtual private voice networks.

PRIVATE VOICE NETWORKS
In private voice networks, the user establishes his own networking topology using private lines or TDM-derived channels. Direct routes are established across major cross-sections while tandem PBXs are defined to provide connectivity among all sites. Uniform dialing and feature transparency are supported by inter-PBX signaling and by administering network routing tables in each PBX.

VIRTUAL PRIVATE VOICE NETWORKS
In virtual private voice networks, each PBX is connected to the public switched telephone network (PSTN), which interprets the signaling received from the source PBX to route the call to the destination PBX. In this case, the routing tables are managed on a centralized basis as part of the carrier virtual private network (VPN) service. Even when virtual private voice networks are implemented, direct traffic between major corporate sites can still be handled over private networks without complicating routing table administration.

Interconnection between the PBX and the packet network is achieved by some sort of gateway or adaptation function.

A typical configuration is a PBX connected to an Enterprise Network Switch, a voice FRAD, or a voice over IP gateway to support connectivity over an ATM, frame relay, or IP network, respectively.

Two modes of operation exist for interconnecting PBXs over packet networks, one analogous to private voice network leased lines and the other analogous to VPN operation. In the former case, the packet network emulates a set of PBX–PBX voice trunks on a preconfigured basis; in the latter, the packet network gateway interprets the signaling from the PBX and dynamically switches the call to the destination. In either case, feature transparency must be maintained.

With dynamic switching, the number of PBX interfaces is determined by the total peak load, not by the number of trunk groups to remote PBXs. Voice routing tables are managed on a logically centralized basis as is the case with virtual private voice networks. Routes can be calculated dynamically on a per call basis, based on the current network conditions, thus simplifying network design. A single update for the whole network is sufficient to add or remove a PBX in the network.

VOICE NETWORKING REQUIREMENTS
Users have all experienced the impacts of noisy lines and excessive delays on the effectiveness of voice communications. In video conferencing, it is a well established fact that while deterioration in video quality can be tolerated, voice quality has to be maintained. Generally speaking, class of service (COS) requirements for voice include low latency (tens of ms is great, 100 ms is asking for trouble, a quarter of a second is highly disruptive) and low delay variations, and low loss for fax, modem, and digital data.

Packet voice gateways are a critical network component in delivering high-quality packet voice networking. Equally important are the attributes of underlying packet network infrastructure.

Gateways have to support the appropriate interfaces and functionality for the given voice environment. A signaling method must be provided to allow the end device to implicitly or explicitly signal application COS needs (e.g., voice versus fax versus modem traffic). Signaling of COS requirements can be done on a per port basis for preconfigured PBX–PBX connectivity, or on a dynamic basis by interpreting signaling from the PBX for dynamic connectivity. Ability to signal rejection of the request based on network resources must also be provided, so that alternate paths can be taken (e.g., via the PSTN).

In all cases, there are a number of fairly complex product implementation challenges associated with supporting preconfigured connectivity of PBX trunks or dynamic (on-demand) switched operation. These include support for analog and digital interface standards to the voice world and a great diversity of signaling standards required for the global marketplace. State-of-the-art dynamic voice compression and fax demodulation are required to ensure voice quality is met and price and performance are optimized.

More specifically, networking performance requirements for supporting real-time voice calls over packet networks (fax is much less of an issue) fall into two categories: control of latency and delay variation and low packet loss.

LATENCY/DELAY VARIATION
Delays much above 70 ms start to impact voice performance. Mechanisms must be provided in the network to limit the maximum delay and delay variations (also called jitter) experienced by voice packets, particularly under congestion conditions caused by bursty high-speed data traffic. These include priority queuing schemes and controlling packet sizes and COS-based routing. Mechanisms must also be provided in the gateways to compensate for these delays (e.g., using dejittering buffers). Packet network delays are due to:

• The time it takes to buffer the packet before transmission (queuing delays are the dominant factor in packet networks and increase with traffic load and with the number of hops).
• The time it takes to assemble a voice packet at the gateway or PC (packetization delay of a few ms is not uncommon).
• The time it takes to transmit the packet over the line at some bit rate and near the speed of light (e.g., transmission delays of tens of ms are not uncommon).

PACKET LOSS
While packetized voice does not represent a high-bandwidth application (unless it is part of a high-resolution video application), reasonable quality can only be maintained if the packet loss rate is kept reasonably low. Packet loss can result if there are line errors, if transmission queues overflow under heavy loads, or if dejittering buffers overflow (due to excessive network delays). Unlike TCP/IP, there are no mechanisms (nor time) to retransmit voice packets in case of loss.

SO, WHAT’S THE REALITY?
In real-world enterprise networks, voice over ATM is the only generally deployable capability. Achieving good voice quality over frame relay and IP will be highly dependent on the attributes of the underlying packet network. For example, a frame relay network engineered for voice and supporting switched virtual circuit (SVC) operation, multimedia classes of services, and local frame fragmentation can be a good basis for an integrated voice/data solution. On the other hand, a frame relay or IP network engineered and designed for data would only be a good solution for packet voice under lightly loaded conditions. In general, the connection orientation of frame relay and ATM networks makes these technologies a better fit with voice requirements. In addition, since many router networks use frame relay or ATM as their underlying infrastructure, bandwidth overhead can be minimized by directly mapping voice onto frame relay or ATM, rather than incurring the additional overhead of IP (the IP tax).

Enterprise users are also looking to frame relay with multimedia classes of service to extend their networks to remote branch locations. A healthcare institution, for example, is running voice over frame relay to their clinics in New York state. Service providers are also aggressively pursuing voice over frame relay. A leader in this area is Infonet, a global service provider, with their Integrated Media Services, which is built on what Infonet calls Multi Media Cell Technology (MMCT) — which itself is based on frame relay SVC class of service capabilities. I estimate that the usage of voice over frame relay is in the same order of magnitude as voice over ATM as measured on a calls/day basis.

Carrying voice over their branch router networks is a new opportunity that is being pursued selectively by some users. At this early stage in the lifecycle of voice over IP, voice over IP gateways are recommended for use only in controlled environments. In one deployment, a Texas bank has provided gateways at eight branches and at the head office using their router network for economic transport. Internal voice traffic is supported as is local access for their customers for on-line banking.

A FINAL NOTE
In conclusion, packet voice can leverage the inherently bursty nature of voice traffic and new compression schemes, and can deliver toll quality voice with much less bandwidth than the traditional 64 Kbps stream. The result is voice and data integration, lower cost per minute for voice traffic, and improved performance for data.

Tony Rybczinski is director of strategic technologies and marketing for Nortel’s (Northern Telecom) newly formed Enterprise Data Networks business unit. Enterprise Data Networks will focus on delivering high-performance data networks globally. The business unit will broaden customer choice by offering new alternatives to increasingly complex data network infrastructures through direct and indirect sales channels. By expanding Nortel’s already broad portfolio of open-standards based products and technologies, Enterprise Data Networks will specifically target opportunities in high-performance data networking. For more information, visit the company’s Web site at www.nortel.com.  E-mail questions or comments to the author at tony.rybczynski@nortel.com.