June 1998

BY TODD KRAUTKREMER

If voice over frame relay (VoFR) were simply a cool technology without a compelling application, it would have long ago joined the ranks of such homeless technologies as the wireless PBX. But VoFR enjoys perhaps the most compelling of all applications: exceptional economics. Reasonable deployment and low maintenance costs coupled with dynamic functionality make VoFR a logical conclusion for organizations looking to improve their existing networks.

VoFR MARKET MOTIVATORS
Although VoFR is almost mystical in function, the economics behind its growing popularity are very real. Of the numerous VoFR market motivators, three are most compelling.

Alternative To International PSTN Services
For quite some time, rates for international public switched telephone network (PSTN) services have far outstripped their U.S. counterparts. In-country PSTN rates can commonly run $0.25/minute with inter-country rates running significantly higher. VoFR implementations using private, public, or hybrid networks can cost as little as $0.01/minute, depending on topology.

Smaller Companies With Limited Purchasing Power
In the United States, voice networking is largely a solved problem for many corporations. Competition among voice carriers has created an environment of diverse public "virtual private network" solutions that offer cost-effective alternatives to private networks. Although these networks can cost as little as $0.05-$0.10/minute, they often require substantial volume commitments that make them unattainable to many small- to medium-sized companies. VoFR offers a public networking alternative for voice that is well suited to these smaller entities.

Migration From Time-Division Multiplexer Networks
The market for frame relay equipment and services is expected to top $14 billion by the year 2000. A substantial part of the growth has been the migration of data from private Time-Division Multiplexer (TDM) networks. Prior to frame relay, TDM was the predominant backbone technology for corporate wide-area networks (WANs) in the United States and abroad, employing switches from the likes of NET, Timeplex, and others. These networks often converged voice and data trunks by muxing them into channelized T1 services. In fact, many early TDM deployments were actually fueled by voice requirements rather than data. Over the years, that relationship flip-flopped as data became the driving economic force.

Frame relay is most cost-effective for geographically dispersed WANs, the same topologies commonly supported by TDM networks. As the economic advantages of frame relay have increased in recent years, so too has the migration of data from TDM backbones to frame relay, leaving voice traffic alone to foot the bill. With the departure of data, the total cost of ownership per Kbps of traffic has risen sharply. This increase is due to the high degree of fixed costs tied up in depreciation expense, annual software licenses, maintenance, staff, and other expenses. As a result, there is significant pressure to migrate voice traffic to public network services. VoFR is an attractive option in that it facilitates low-cost local loop consolidation of voice and data, especially for lower-speed 56/64 Kbps trunks.

CURRENT MARKET
The VoFR market is growing at a steady clip, with equipment and service revenues expected to exceed $100 million by the year 2000. Recent studies in the U.S. have found that more than 30 percent of surveyed corporations ranked VoFR support as a "very important" factor in their decisions regarding branch internetworking equipment and services. Worldwide public VoFR ports are expected to grow from just 8,338 ports in 1996 to more than 20,000 by the year 2000, with an additional 90,000 private VoFR ports by that same time.

As an industry, VoFR is clearly moving from the visionary deployment phase to that of the early adopters. Increased standardization, strong vendor support, and the early signs of carrier acceptance (especially in international markets) fuel this move. The Frame Relay Forum has ratified VoFR-supporting specifications for segmentation (FRF.11), as well as framing and call processing (FRF.12). Vendor support is growing rapidly, with all of the leading frame relay access vendors -- including Cisco, Motorola, Hypercom, and Sync Research -- offering VoFR-capable frame relay access devices (FRADs).

On the carrier side, VoFR support has been slower to develop, most likely due to concern about the erosion of their public voice networking empires. However, there are signs that the major U.S. carriers are starting to embrace VoFR. Ameritech publicly announced support in late 1997; Sprint and MCI are expected to follow suit in 1998. Value-added network providers and those with a large base of multinational connections -- like InfoNet -- were early on the VoFR bandwagon and position it as a key service differentiator. Major international carriers are following a similarly cautious approach as their U.S. counterparts, leaving initial VoFR service offerings to the burgeoning number of emerging value-added networks and competitive local exchange companies (CLECs).

VoFR TOPOLOGIES AND REQUIREMENTS
Indicative of voice networking in general, there are many possible VoFR topologies. However, there are two particular topologies that are most common: point-to-point and hierarchical.

Point-To-Point
The first deployments of VoFR technology provided integrated voice and data connectivity in a point-to-point fashion between two offices. Early VoFR vendors -- most notably MICOM, now part of Nortel -- pioneered proprietary schemes for replacing analog "plain old telephone service," or POTS, lines by packetized voice (and facsimile) traffic and multiplexing it with data, first across leased lines and later frame relay. Today, point-to-point implementations tend to be deployed as more of a quick fix rather than as a comprehensive networking solution. These implementations are motivated almost entirely by economics rather than quality of service (QoS), with a return on investment of three to six months being commonplace. Due to less sophisticated voice and data networking requirements and the simplicity of point-to-point topologies, these implementations generally have limited connectivity and multiprotocol handling requirements. Issues like management, bandwidth efficiency, and voice quality take a backseat to realized cost savings.

Hierarchical
At the other end of the spectrum from point-to-point are hierarchical topologies. These topologies are typified by many remote sites or branches being interconnected to one or more central site locations via frame relay permanent virtual circuits (PVCs). Companies that conduct business through a remote branch framework -- such as banks, retail stores, and insurance agencies -- are very dependent on voice and data communications to drive their revenue generation process. Because of the higher telecommunications costs associated with being geographically dispersed, most corporations have already deployed a voice network solution using either a private TDM or a public virtual private network solution. For these organizations, VoFR represents an incremental reduction in voice costs, especially if they have already migrated their branch data traffic to frame relay.

Since the branch voice network is often intertwined with a company's revenue generation functions, the requirements for VoFR support in hierarchical topologies is fundamentally different than that of simple point-to-point connections. To start with, hierarchical topologies have many remote branches interconnected to a few centralized locations - usually corporate headquarters or data centers. Equipping this topology with VoFR support requires smaller, cost-effective voice-capable FRADs or routers at the branch sites, usually supporting up to four analog or digital voice ports, and large-scale central site platforms supporting potentially hundreds of PVCs and numerous voice ports connecting to a central PBX.

In addition to branch-to-central site calls, many branch networks demand branch-to-branch calling. This means the VoFR solution must be capable of routing calls up and down branch PVCs. Unlike many point-to-point environments, QoS is a significant issue within business-critical branch networks. Although saving money is important, it is secondary to maintaining reliable and high-fidelity voice connections. As with voice, the data requirements of branch topologies are more demanding, with many of sites requiring support for a broad range of legacy and LAN (local area network) protocols, such as SNA, BSC, IP, and IPX.

TODAY'S VoFR SOLUTIONS
VoFR technology has rapidly evolved in the last several years. Today, a growing number of vendors are delivering VoFR solutions that target the requirements of complex, hierarchical branch networks. Although VoFR standards are evolving at a rapid pace, achieving the challenging blend of voice reliability, fidelity and bandwidth efficiency goes beyond such features as echo cancellation, and requires a healthy dose of innovative features and capabilities.

Quality Of Service
In order to converge voice and data traffic onto a single frame relay trunk, voice-capable FRADs must employ QoS features that rival the functions of ATM. Mixing delay and delivery rate-sensitive voice traffic with "bursty" data requires complex queuing algorithms that can process frames not just by traffic priority but also based on the maximum elapsed time in queue. Dynamic bandwidth allocation and shaping mechanisms must engage within milliseconds to transmit pending voice traffic and reduce the impact of jitter (erratic delivery of successive voice samples).

Voice Compression
Over the years, voice compression technology has seen constant improvement. Today, VoFR standards utilize a number of VoFR compression techniques that range from 8 to 32 Kbps per voice channel. Some VoFR equipment companies, such as Nuera Communications, can provide toll quality compression at 4.8 Kbps. Such high compression capabilities allow numerous voice channels to be transmitted in the same bandwidth that it once took to support a single channel.

Frame Relay Segmentation
Since queuing and transmission latency in both FRADs and switches are directly dependent on frame size, mixing frames of widely varying sizes will provide variable latency (jitter) for voice transmissions. Frame relay segmentation alleviates this problem by segmenting all voice and data transmission to a fixed frame or "cell" size. The advantage over ATM is that a frame relay "cell" can be significantly larger than ATM 53-byte format, which allows the user to configure the network in more data bias (larger frame size) if so desired.

Frame Packing
Most digital signal processors (DSPs) used in VoFR implementation compress and digitize voice into small samples -- usually around 24 bytes. These samples are then placed in frame relay packets and transmitted. However, segmenting all voice and data traffic to just 24 bytes can dramatically impact the performance of data. By taking multiple voice samples (usually three to five, depending on trunk speed) and packing them into a single frame, larger frame sizes can be used to improve data performance without interfering with DSP efficiency or voice quality.

Jitter Compensation
As hard as the FRAD tries, a certain amount of jitter is unavoidable in packetized voice transmission due to switching latencies, congestion, and lost frames. To compensate for this, the receiving FRAD employs a jitter buffer that resynchronizes frames coming off the WAN before streaming them to the receiving voice stations (telephone, PBX, etc.).

Silence Suppression
The human speech pattern is filled with periods of silence. In a digital sense, silence is construed as a discernable element and would normally be transmitted. However, almost all of the DSPs used in today's VoFR implementations have the ability to detect and weed out silence patterns, which reduces bandwidth.

Lost Speech Interpolation
One problem affecting frame relay networks is that frames occasionally get discarded. From time to time, any number of situations, such as congestion, excessive bursting, or PVC rerouting, can cause the loss of one or more frames. When transmitting compressed voice, one or more lost frames (depending on frame size) can cause distorted speech. This is commonly referred to as clipping. Anyone with a digital cellular phone has at one time experienced the annoying clipping of conversations that occurs as people approach a fringe coverage area or switch between cells.

To minimize clipping caused by frame loss, some of the more advanced VoFR FRADs employ a lost speech interpolation technique to blend over the lost speech pattern. This technique attempts to recreate the lost frame(s) by taking the last couple of milliseconds of speech pattern from the preceding frame and the first couple of milliseconds of pattern from the following frame and algorithmically interpolating the missing speech pattern.

On-Network Call Switching
As previously discussed, many branch networks require branch-to-branch communications to support the company's workflow. There are three approaches to providing this support. The network can be designed with PVCs that interconnect all branches. In all but the smallest of networks, this approach quickly becomes cost-prohibitive. Secondly, all calls can be routed through the central PBX at the hub of the network. Although very cost-effective, the multiple voice compression and decompression cycles required to come off the frame relay network, go through the PBX, and return onto the frame relay network can significantly impact voice quality. The third approach is for the central "hub" FRADs to provide on-network switching of voice calls. In this case, the FRAD acts as an extension of the central PBX and routes calls between branches without intermediary compression/decompression steps.

Voice Management
As it is with data networking, one cannot achieve high availability and deterministic performance in voice networking without extensive management. However, cohabitation with data exacerbates voice network management in a VoFR environment. While channel-oriented voice management works well in TDM and public network solutions, a packet-oriented strategy is required for VoFR support, especially if voice and data are sharing the same PVC. In order to understand the performance of voice traffic end-to-end, network managers need to understand how voice and data traffic are contending for shared network resources. They also need to monitor service-affecting metrics like latency, congestion, frame delivery rate (measure of discarded frames), and call clearing conditions.

The good news is that a new generation of frame relay management technology is poised to make VoFR management easier. Sync Research and other leading vendors have recently announced enhancements in frame relay access probe (FRAP) WAN monitoring probes and associated management software that incorporates extensive VoFR performance and service-level management capabilities. Providing concurrent voice and data management will enable network managers to proactively capacity-plan their frame relay networks in response to ever-changing traffic requirements, and it equips them with the diagnostic tools necessary to rapidly isolate problems and restore service levels. It also will enable easy accounting of voice traffic for bill-back purposes.

SUMMARY
The voice over frame relay market is enjoying rapid growth throughout the world, especially in countries where voice networking is still quite expensive. For many applications, compelling economics fuels the migration to VoFR, with return on investments of less than six months being commonplace. Point-to-point topologies have been the predominate implementations to date and are easily supported by today's VoFR FRADs and routers. However, supporting business-critical branch networks and their more complex hierarchical topologies requires much more sophisticated VoFR solutions, with value-added features that insure reliability, scalability, efficiency, and deterministic voice quality across a broad range of operating conditions. The remaining challenge for VoFR is management. A new generation of frame relay management technology is well positioned to deliver on the promise of performance and service-level management for integrated voice and data networks.

Todd Krautkremer is vice president of strategic marketing at Sync Research. Sync has played a key role pioneering several industry-shaping technologies, including SDLC-to-LLC2 conversion, the industry's first frame relay access device (FRAD), and frame relay service-level management technology. Sync is a global leader in WAN access and management solutions that make it safe for corporations to migrate their business-critical applications from private leased-lines to PSVN services. By leveraging inherent switching, quality of service, and multiprotocol transport capabilities, Sync's innovative products unleash the power of PSVN services to deliver cost and productivity advantages that surpass today's expensive and complex SNA and router-based WANs. For more information, contact the company at 714-588-2070 or visit their Web site at www.sync.com.

Choosing Frame Relay For Voice

With an expanding variety of choices for integrating voice, fax, and data over wide-area networks (WANs) comes uncertainty about how to make valid comparisons between the alternatives. A successful network design is one that controls communication costs while supporting an increasing number of users, ever more bandwidth, and a growing number of points of presence, without sacrificing quality or reliability. Network designers must evaluate voice over frame relay and IP networks for delay and efficiency, and the ability to scale and migrate to larger networks in the future. Ultimately, the resulting quality of service will determine if the network is one that the end user will choose to use.

DELAY
Delay is the quality factor that users notice most. As with ambient temperature and noise levels, variances in delay within users' tolerance ranges pass unnoticed. End-to-end delay is what matters to the user. This includes delay from all sources in the transmission path: access equipment input and output buffers, compression processes, serialization delay, and network delay.

For voice traffic, compression/ decompression and serialization of voice signals at rates around 8 Kbps typically add delays of over 60 milliseconds, so a few additional milliseconds aren't critical, or even audible. End-to-end delays of 130 milliseconds are generally not distinguishable from the toll-quality of the PSTN in most cases, so a network which adds 50-70 milliseconds of delay to that of a low-delay voice FRAD will result in an application which compares favorably with the PSTN.

EFFICIENCY
Efficiency is measured as payload as a percentage of the total bandwidth available. As opposed to constant bit rate services such as TDM and ATM AAL-1, IP and frame relay both provide efficient variable-length packets which ensure that only fully-packed frames or packets are sent. Frame relay headers are only 5 to 7 bytes -- with typical payloads, the overhead can be as low as about 5 percent. For IP traffic with 20-42 bytes, however, the overhead can be 25 percent to 40 percent.

While users notice delays, efficiency is something that hits carriers hardest. Users need to move a certain number of bytes of payload from certain sources to certain destinations. The less efficient a carrier's network, the more bandwidth is required to allow the users to move those bytes. This leaves the carrier with several choices: buy more bandwidth, which costs more money, or use the same bandwidth to accommodate fewer users and make less money. The fact that neither of these options seems palatable explains why efficiency is critical to a carrier in a competitive world, and why efficiency remains a key consideration in network design.

MIGRATION
Most IP traffic in the world is carried over frame relay or ATM (layer 2) networks. Many large carriers put in an ATM core network and provide ATM/frame relay interworking at the edge of their networks. Smaller carriers typically put in a frame relay network -- when their traffic grows to very high levels, they migrate to an ATM backbone network. By taking this two-step approach, the high cost of ATM equipment and the high cost of very high-speed transmission systems between the switches are deferred until the traffic level can justify the cost. In any case, IP, frame relay, and ATM can be and already are successfully interwoven into a seamless WAN fabric worldwide.

QUALITY: THE BOTTOM LINE
In competitive environments, you can't attract the most profitable customers unless you meet their quality requirements. But if you don't serve them efficiently, your costs may run too high to take gross profit to the bottom line. So what do you need to make sure you can still meet quality objectives of multimedia users?

Vendors of frame relay equipment have developed ATM QoS-like features to inject more certainty into delay times transiting a frame network. Likewise, IP equipment vendors have developed RSVP to allow applications to request that bandwidth be reserved across an IP network for the duration of their session. In the case of IP/RSVP, consultant Ron Jeffries quotes a router guru who admits that this approach requires routers "to perform unnatural acts." IP was not designed -- and routers were not built -- to handle steady-state, real-time traffic.

In the case of frame relay, consultant Lynne Nye has suggested that end users and network designers pay attention to "Quality of Design" (QoD). That is, that proper network design to match the backbone and access links will achieve the same result of assured bandwidth and satisfactory delay, but with a lot less complexity.

• Access equipment: Pay attention to quality and delay when choosing access equipment, including voice compression equipment, and you will have won half the battle. It's a lot cheaper to cut delays by buying high-performance access than it is to leave excess capacity in the network to try to minimize delays there. Many carriers have solved "network delay" complaints by recommending or providing low-delay access devices to their customers, and the problems have disappeared without any change to the network.
• Backbone network design: Knowing what demands users will place on the network allows you to meet their demands at a minimal cost. Matching desired delays with acceptable efficiencies will allow you to determine the most profitable approach.
• Configuration and planning: Proper configuration of the access equipment according to user needs, and proper planning of the service parameters to meet actual use, are the foundation of good performance.

With these elements in place, adequate performance -- and satisfied users -- can be assured quality will be achieved at efficiencies greatly exceeding PSTN applications, leading to lower costs and improved profits.

Andy Voss is vice president of marketing and Jim Murphy is the OEM product manager for Nuera Communications. Nuera manufactures the award-winning Access Plus series of products for voice compression over frame relay, TCP/IP, satellite links and/or leased lines. The F200IP IP gateway features voice and data bandwidth management and full call switching between networks that allow managers to select call routing paths to reduce operating costs and to improve communications quality. Applications include carrier and corporate domestic and international voice/fax services over packet networks. For more information, contact Andrew Voss at 619-625-2400, ext. 1259, or visit the company's Web site at www.nuera.com.