The Case for Cooperative Networking
by Gil Biran,
Vice President of Research and Development,
RAD Data Communications Ltd.
Table of Contents
There’s a lot of talk about the suitability
of Frame Relay, IP and ATM for carrying voice and the advantages
of one over the other. It is clear that there are advocates
of all three technologies and that each is suited to particular
needs and environments. That being the case, they will all probably
be around for some time to come. It is therefore becoming apparent
that there will be a need for interworking between them.
In this paper we will discuss the voice-enabling
mechanisms employed by all three technologies and solutions
The Trend toward the
Integrated Voice and Data Network
In the last few years, data networks have
been growing at a much faster rate than voice networks, mainly
due to the growth of the Internet. Soon the amount of data traffic
will exceed that of voice traffic. As a result of this trend,
more and more voice is being sent over data networks (Voice
over Frame Relay, Voice over IP and Voice over ATM) than data
is being sent over voice networks (via V.34 and V.90 modems).
When Frame Relay was introduced in the early
1990s, the data technology was not originally designed to carry
voice. Despite valid reservations about the reliability of voice
over frames, the promise of “free voice” eventually
proved too alluring. Soon users were experimenting with transporting
voice over their Frame Relay devices while equipment vendors
worked overtime to make the promise of quality voice over Frame
Relay (VoFR) a reality.
As the public Internet exploded in the mid-1990s
and users began implementing IP-based networks, the call for
voice over IP (VoIP) grew louder. Here, too, equipment manufacturers
are developing products to enable inexpensive, universal voice
over data networks.
Carriers, however, were caught in a dilemma.
Could they afford to cannibalize their highly profitable public
switched telephone network? Could they not afford to capitalize
on the demand for digital voice? The drama is just unfolding.
Although significant progress has been made
in engineering packet networks (Frame Relay, IP and ATM) to
carry voice as well as data, today’s market is demanding a true
convergence of these technologies into a single and ubiquitous
communications service without being limited by the underlying
technology. The next challenge, then, is to develop interconnection
and interworking standards in order to deliver voice services
ubiquitously over Frame Relay, IP and ATM.
The Nature of the Data Network and its Implications
The Nature of the Data Network
Frame Relay, IP and ATM are known as
packet or cell switching technologies. This is in contrast to
the public telephone network, which is a circuit switching technology,
designed to carry voice transmissions. Frame Relay and IP insert
data into variable-sized frames or packets. ATM chops data into
small cells, which facilitates fast switching of data through
The packet switching and cell switching networks
perform statistical multiplexing. That is, they dynamically
allocate bandwidth to various links based on their transmission
activity. Since bandwidth is not reserved for any specific path,
the available bandwidth is allotted according to network needs
at any particular time.
Compare this to the traditional voice (or
circuit switching) network, in which a path is dedicated to
the transmission for the duration of the call, which is sent
in a continuous bit stream. The line is monopolized by a call
until it is terminated, even when the caller is put on hold
and during periods of silence. Although this guarantees reliable
and immediate transmission of voice, it results in very inefficient
use of bandwidth. A line that is dedicated to the telephone
cannot be utilized by other data even when there are no voice
Originally designed to handle bursty data
traffic, packet switching networks (except for ATM) are inherently
less efficient than the circuit switching network in dealing
with voice. To achieve good voice quality, the delay of voice
packets across the network must be minimal and fixed. Due to
the shared nature of the packet/cell switching network, it might
take time for transmissions to travel across the network. A
transmission can be delayed because of network congestion. For
example, it might “get stuck” behind a long data transmission
that delays other packets. Network congestion can also result
in dropped packets, which also detrimentally affects the integrity
of voice transmissions.
Voice-Enabling the Data Network
Unlike most data applications,
voice is very sensitive to delay. Good voice quality provides
a faithful recreation of the conversation, with the same tone,
inflection, pauses and intonation used by the speakers. Long
and variable delays between packets result in unnatural speech
and interfere with the conversation. Dropped packets result
in clipped speech and poor voice quality. Fax transmissions
are even more sensitive to the quality of the transmission and
are less tolerant of dropped packets than voice.
One way to deal with the problem of delay
and congestion is to add bandwidth to the network at critical
junctures. Although this is feasible in the backbone, it is
a costly and ineffective solution in the access arena, defeating
the “bandwidth sharing” benefits of packet networks.
The best solution is to implement mechanisms at the customer
premises, access node and backbone which manage congestion and
delay – without increasing bandwidth – such as setting priorities
for different types of traffic. Therefore, smart access equipment
was developed, that could implement procedures to reduce network
congestion and the delay of voice packets without adding bandwidth.
over Frame Relay (VoFR)
Of the three popular packet/cell technologies
(Frame Relay, IP and ATM), Frame Relay is the most widely deployed.
Frame Relay is commonly used in corporate data networks due
to its flexible bandwidth, widespread accessibility, support
of a diverse traffic mix and technological maturity.
Frame Relay service is based on Permanent
Virtual Connections (PVCs). Frame Relay is appropriate for closed
user groups and is also recommended for star topologies and
when performance needs to be predictable. VoFR is a logical
progression for corporations already running data over Frame
Voice Frame Relay access devices (VFRADs),
such as RAD’s MAXcess integrated bandwidth manager, integrate
voice into the data network by connecting the router (or using
the integrated router available on certain MAXcess models),
SNA controller and the PBX at each site in the corporate network
to the Frame Relay network.
Many VFRADs, such as RAD’s MAXcess, employ
sophisticated techniques to overcome the limitations of transporting
voice over the Frame Relay network without the need to add costly
These techniques are discussed below for Frame Relay. Implementation
for IP and ATM is discussed in the Voice over IP (VoIP) and
Voice over ATM (VoATM) sections, respectively.
The VFRADs’ prioritization schemes
“tag” different applications according to their sensitivity
to delay, assigning higher priority to voice and other time-sensitive
data such as SNA. The VFRADs let the higher priority voice packets
go first, keeping the data packets waiting. This has no negative
effect on data traffic, as voice transmissions are relatively
short and, being compressed, require very little bandwidth.
They can therefore slip into the data network alongside the
heavy graphics, payroll information, e-mail and other data traffic
without perceptibly encumbering overall network performance.
Frame Relay service providers have also begun
to offer different Quality of Service (QoS). Users can purchase
the highest quality of service, Real-Time Variable Frame Rate,
for voice and SNA traffic. The second quality Frame Relay service,
Non-Real Time Variable Frame, is typically purchased for LAN-to-LAN
and business class Internet and intranet traffic. The lowest
quality of service, Available/Unspecified Frame Rate, is used
for e-mail, file transfer and residential Internet traffic.
In addition, the VFRAD can be configured to assign less sensitive
traffic with a Discard Eligibility (DE) bit. These frames will
be dropped first in case of network congestion.
The MAXcess and other VFRADs
incorporate fragmentation schemes to improve performance. Data
packets are divided into small fragments, allowing higher priority
voice packets to receive the right-of-way without waiting for
the end of long data transmissions. The remaining data packets
in the data stream are momentarily halted until the voice transmission
The down-side of fragmentation is that it
increases the number of data frames, thereby increasing the
number of flags and headers. This increases overhead and reduces
bandwidth efficiency. RAD’s FR+ application provides an enhanced
fragmentation mechanism which fragments data frames only in
cases where voice packets arrive at the switch during a data
transmission. Otherwise, the long data frames are sent intact.
Controlling Variable Delay
Variation in the arrival times
between packets, also called jitter, causes unnatural-sounding
voice instead of a smooth voice stream. If a packet does not
arrive in time to fit into the voice stream, the previous packet
is replayed. This can seriously detract from voice quality.
To avoid the effect of jitter, VFRADs such as the MAXcess detain
each packet in a jitter buffer, giving subsequent packets time
to arrive and still fit into a natural voice flow. Since the
jitter buffer adds to the overall delay of voice transmissions,
the optimal jitter buffer should fit the network’s differential
delay. Better access devices, like RAD’s MAXcess, employ adaptive
jitter buffering, which continuously monitors the network delay
and adjusts the queuing period accordingly.
Voice compression allows the
packet switching network to most effectively carry a combination
of voice and data sessions without compromising voice quality.
Since Frame Relay access is usually at data rates of 56/64 kbps,
low bit-rate voice compression algorithms such as ITU G.723.1
and G.729A permit the greatest number of simultaneous multiple
calls while maintaining high quality voice. Vendors such as
RAD, which have implemented voice compression algorithms in
their Frame Relay access devices, can offer greater bandwidth
savings, reduced network congestion, and high quality voice
In a telephone conversation,
only about 50% of the full duplex connection is used at any
given time. This is because, generally, only one person talks
while the other person listens. In addition, voice packets are
not sent during interword pauses and natural pauses in the conversation,
reducing the required bandwidth by another 10%. Silence suppression
frees this 60% of bandwidth on the full duplex link for other
voice or data transmissions.
Echo cancellation improves the
quality of voice transmissions. It eliminates the echo that
results from the reflection of the telephony signal back to
the caller, which can occur in a 4-wire to 2-wire hybrid connection
between the VFRAD and the telephones or PBX. The longer it takes
the signals to return to the caller, the more perceptible the
Voice over IP (VoIP)
Internet Protocol (IP) is a connectionless
protocol in which packets can take different paths between the
endpoints and all paths are shared by packets from different
transmissions. This enables efficient allocation of network
resources, as packets are routed on the paths with the least
congestion. Header information makes sure that the packets reach
their intended destinations and helps reconstruct the messages
at the receiving end. To ensure QoS, however, all packets should
use the same path. IP headers are large (20 bytes) as compared
to the headers of Frame Relay frames (2 bytes) and of ATM cells
(5 bytes). Headers therefore add a lot of overhead to IP traffic.
IP networks employ the same types of bandwidth-saving
schemes as the Frame Relay network, including fragmentation,
jitter buffering, prioritization, voice compression, silence
suppression and echo canceling.
techniques used for VoIP are
different from those employed by Frame Relay access devices.
Prioritization is directly related to QoS. The key IP QoS protocol
was RSVP, which allowed the sender to request a certain set
of traffic-handling characteristics for traffic flow, but was
not widely adopted. Today, the Intserv working group of the
IETF is developing a simpler, more promising solution. The Differentiated
Services Model uses the Type of Service (ToS) octet field of
the IP header to classify traffic at the borders between the
customer and service provider or Internet service providers
(ISPs). Currently, there is still no viable QoS for IP services.
is performed in a similar fashion as Frame Relay fragmentation.
Fragmentation, although required to reduce the overall delay
of voice traffic, adds a lot of overhead to IP transmissions
due to the large size of IP headers. IP voice traffic therefore
consumes 50% more WAN bandwidth than Frame Relay voice traffic.
However, as IP matures, header compression and improved routers
will eliminate these shortcomings.
is vital in Voice over IP because traffic usually travels over
low-speed links. Some small and medium-sized enterprises, for
example, may be connected to the virtual private network (VPN)
at only 28.8 kbps. Microsoft Netmeeting, for example, a popular
voice application for PCs and laptops, supports ITU G.723.1
voice compression for transmission over dial-up modems. The
ITU G.723.1 standard for voice compression over IP ensures toll
Jitter buffer, silence suppression
and echo cancellation
techniques are similar to those employed in VoFR. Echo cancellation
is extremely important in VoIP, which often suffers from long
Voice over ATM (VoATM)
Asynchronous Transfer Mode, or ATM, is a
multiservice, high speed, scalable technology. It is a dominant
switching fabric in carrier backbones, supporting services with
different transfer characteristics. ATM simultaneously transports
voice, data, graphics and video at very high speeds. On the
down side, ATM services are expensive and not yet universally
Large enterprises are increasingly connecting
headquarters and main offices to the wide area network via broadband
links such as ATM to accommodate their vast amounts of voice
and data transmissions, such as heavy graphics, payroll information
and voice and video conferencing.
is built into ATM, with its small, fixed-size, 53-byte cells.
Very fast ATM switches speed data through the ATM network. The
high bandwidth associated with ATM reduces congestion problems,
providing extremely reliable service. Carriers can therefore
promise customers Quality of Service (QoS), stipulated in Service
Level Agreements (SLAs).
is implemented through QoS parameters.
ATM was designed from the outset to carry voice as well as all
types of data. ATM Adaptation Layer 1 (AAL1) protocol in ATM’s
Constant Bit Rate (CBR) service was the de facto standard for
VoATM. However, this protocol proved inefficient for voice applications.
CBR, the highest quality class of ATM service, provides Circuit
Emulation Service (CES), which transmits a continuous bit stream
of information. This allocates a constant amount of bandwidth
to a connection for the duration of a transmission. Although
it guarantees high quality voice, CES monopolizes bandwidth
that could be used for other applications. In addition, in the
interest of reducing delay, CES might send the fixed-size ATM
cells half empty rather than waiting 6 milliseconds for 47 bytes
of voice to fill the cell. This wastes over 20 bytes of bandwidth
per ATM cell. Dynamic Bandwidth Circuit Emulation Service (DBCE
S) is a variation of CES. DBCES does not send a c onstant bit
stream of cells, but transmits only when there is an active
voice call (off hook). However, like in CES, the cells might
remain partially empty. Therefore, using AAL1 for VoATM increases
the overhead of voice transmissions and wastes bandwidth.
AAL2’s Variable Bit Rate (VBR-RT) service,
as specified in ITU-T recommendation I.363.2, emerged as the
standard of choice for VoATM. The structure of AAL2 allows for
the packing of short packets (1 to 45/64 bytes), also called
minicells, into one or more ATM cells. (This resembles Frame
Relay’s and IP’s variable sized fragments.) In contrast to AAL1,
which has a fixed payload, AAL2 enables a variable payload within
cells. This functionality provides a dramatic improvement in
bandwidth efficiency over structured or unstructured circuit
emulation using AAL1. In addition, AAL2 supports voice compression
and silence suppression and allows multiple voice channels with
varying bandwidth on a single ATM connection.
is not necessary in pure-ATM
networks, which enjoy ample bandwidth. However, in hybrid ATM-Frame
Relay networks (for example, with ATM headquarters and Frame
Relay branches), voice compression is required since Frame Relay
uses voice compression. ATM must therefore be equipped to support
voice compression that will work with VoFR equipment at the
Standards with Room for Interpretation
The interworked voice and data network of
the future promises the best of all network worlds: the installed
base of Frame Relay, the speed and quality of ATM and the ubiquity
of IP. Currently, fragmentation techniques in Frame Relay, IP
and ATM are quite similar, but prioritization techniques, signaling
protocols and voice compression algorithms are not compatible.
Progress is being made toward standardization within each protocol
and interworking between them, but considerable work remains.
The Frame Relay Forum has defined standards
for sending voice over Frame Relay. Implementation Agreement
FRF.11 presents basic Frame Relay definitions but allows a lot
of leeway regarding voice switching and negotiation between
different VFRADs. This necessitates a proprietary solution,
precluding interoperability between equipment of different vendors.
Phase 2 of FRF.11, which deals with interoperability
issues, is not yet implemented. As long as Frame Relay is based
on Permanent Virtual Connections (PVCs), which establish fixed
links, there is little incentive for vendors to interoperate.
With PVCs, all nodes are known, and same-vendor equipment is
used on all end devices. If Switched Virtual Connections (SVCs)
were more widely used paths would be defined dynamically, and
interoperability would be elevated to a market requirement.
In the meantime, a proprietary interworking method is an acceptable
solution in private corporate networks.
ITU H.323 defines interoperability standards
for voice and multimedia applications over IP. It determines
endpoint negotiation and information format. It does not, however,
address encoding, prioritization or security. And like other
standards, some of the H.323 definitions are subject to interpretation.
Therefore, H.323 does not guarantee interoperability between
equipment of different vendors. Some IP vendors are collaborating
on an interoperability profile based on ITU H.323 and on the
upcoming H.225.0 Annex G standard. Its goal is to achieve interoperability
between gateways and gatekeepers of different vendors, enabling
deployment of different IP platforms at either end of the network.
Once the profile is released, however, extensive testing will
be required to determine the actual level of inter-vendor compatibility.
Variable Bit Rate AAL2 is recommended for
VoATM. Although it sounds promising, the AAL2 standard is not
yet completely developed. It will most likely suffer from the
same inter-vendor interoperability problems as the VoFR and
VoIP solutions, at least in the short run.
The Need for Interworking
Since a comprehensive standard has not been
adopted for any one technology, it is unrealistic to expect
the emergence of interoperability standards between technologies
in the near future. Interworking solutions will therefore have
to be proprietary. It is essential that the interoperability
be transparent to the users, who want to communicate through
the network efficiently and without concern for the technological
Due to the lack of interoperability standards
for voice communications over Frame Relay, IP and ATM, vendors
must develop proprietary interworking solutions.
There are many situations in which interworking
between technologies is required within a corporate network.
For example, corporations that are running data and voice over
Frame Relay might require VoIP to extend the network to remote
locations that don’t have a Frame Relay infrastructure without
deploying additional equipment. This may also be required for
telecommuters working from home, salespeople working from hotel
rooms and resellers that want to access information. Consider
the configuration below, in which IP is extended to hotels or
Interworking between Frame Relay and IP is
vital in this scenario. This is easier said than done, however,
as Frame Relay and IP use different standard voice compression
algorithms (ITU G.729 and ITU G.723A, respectively). There is
also a discrepancy between signaling methods. The FRF.4 recommendation
for switched virtual circuits is not widely used as the basis
for Frame Relay voice switching. In fact, there is no standard
for voice switching in Frame Relay networks. In any case, it
does not interwork with VoIP voice switching based on the H.323
In large corporate networks with many remote
branches using Frame Relay service and VoFR, there is a need
for high speed ATM service in order to support the required
amount of traffic at company headquarters. In this case there
is a need for interworking between Frame Relay and ATM in general
and VoFR and VoATM in particular.
Another reason for the desirability of multiple
voice technology support over the same platform is to enable
migration to different technologies without losing the initial
investment in existing equipment.
RAD is developing a pre-standards strategy
to facilitate interworking between Frame Relay and IP. The strategy
will provide a migration path from Frame Relay to IP technologies,
which can be an important advantage when IP services become
available. RAD is currently developing an interworking solution
between Frame Relay and IP. The VoFR-VoIP product will perform
signaling conversion and negotiate with the remote IP product
in order to choose a common voice compression algorithm and
Without a doubt, the data revolution will
only gain momentum in the coming years, with more and more voice
traffic moving onto data networks. Vendors of voice equipment
will continue to develop integrated voice and data devices based
on packetized technology. Users with ubiquitous voice and data
service integrated over one universal infrastructure will benefit
from true, seamless, transparent interworking between voice
and all types of data. RAD Data Communications will remain in
the forefront of industry efforts to provide universal services.