November  1999

The Evolution To Packet Switching

BY HARVEY KAUFMAN

The continuing rapid growth in both voice and data traffic has acted as a major motivation for converging the two separate networks. A single merged infrastructure holds the promise of both a greatly improved economic model, and the availability of competitive products for the emerging multimedia market. The first major challenge to this end, achieving a reliable approximation of the necessary real-time, constant-bit-rate delivery characteristics, has begun to benefit from developments in data transport. The second major challenge, and the real key to convergence, lies in creating a true counterpart to the highly developed common control systems used in circuit switching.

In the world of telecommunications, a switching system can be any information delivery network in which a member terminal is able to selectively connect with or through any one of a number of other member terminals. The structure of any voice switching system can be viewed as comprising three functional subsystems that, in combination, determine the capability and functionality of that system. These three are:

1. The matrix — a delivery medium for transporting session information among member terminals;
2. The control system — services available to all member terminals for initiating, modifying, managing and reporting on communications sessions; and
3. The terminal interfaces — adapters/converters for making connected devices or network terminations compatible with the matrix and with the control.

CIRCUIT SWITCHING
Manual Control
Manual switches, commonly referred to as cord boards, traditionally used voice modulated (analog) DC voltage on copper wire as the matrix/delivery medium — the same conduit used today to deliver voice between a switch interface and an industry-standard analog telephone. The terminal interfaces in those systems were integral to the standardized telephones that were supplied by the operating telephone company. The access and control mechanisms resided in the operator, who physically manipulated the cord board.

Automated Controls
The first step in automation was to allow member telephones to directly establish a connection to other member terminals without operator intervention. The major advantages associated with the later introduction of microprocessor controls initially had more to do with economics than with enhanced performance. In addition to reducing the amount of control hardware, it greatly reduced the labor content in implementing upgrades, adds, moves, and changes (the average business telephone moves one to two times a year), and led to participation by third-party application suppliers.

Microprocessors also enabled the implementation of a new generation of switching matrices that allowed multiple sessions to time-share a common transmission medium. A number of multiplex schemes were produced commercially, including frequency division multiplex, and various forms of time division multiplex (e.g., pulse amplitude modulation, pulse width modulation, delta modulation, and the now standard pulse code modulation), all of which were circuit-switch implementations. Each provided the guaranteed bandwidth, low transmission delay, and constant delivery rate essential for acceptable voice service.

Operation, Administration, Management, And Provisioning
In normal operation, a computer common control processor unit (CPU) continually scans all member terminals so that it can detect transitory requests for service (e.g., off-hook, hookflash, dialed digit) in real time. The many other functions it has to perform are transaction driven. The CPU is called on to execute programs defined by the detected stimuli without compromising its real-time priorities.

Because circuit switches are deterministic, meaning that each session resides in an assigned location in time and/or space, an established session adds no processing burden to the CPU other than the continued monitoring for a request for change. The various housekeeping and other essential system functions all have to be performed without violating the real-time constraints. All these activities have to be factored in, along with estimated feature activations, when specifying a CPU busy hour call attempt (BHCA) rating.

As circuit-switching systems have grown busier and more sophisticated, it has become common practice to use multiple distributed microprocessors. Offloading of non real-time functions from the main processor helps to maintain application and feature performance levels without having to derate the matrix. A typical 300-line PBX stored program control today might include two to three dozen distributed microprocessors.

Features, Applications, And CTI
A telecommunications switch exists to provide connections that will support communications sessions. Consequently, the various “features” provided by such switches are associated with making, modifying, or reporting on those connections. Some features, such as station camp on, callback queuing, and call forward, are controlled by the station user. Others, like LCR, station hunt, and DNIS, are programmed by the system administrator and are transparent to the station user.

The performance level of features that require computation and/or referral to large databases can fall below acceptable levels due to the demands of real-time priorities when call activity begins to increase. A feature of this nature is commonly considered an “application,” and typically runs on a dedicated server platform in order to avoid the alternative of scaling down the switching system. As circuit-switch manufacturers began to adopt standard control bus interfaces (e.g., CSTA, TAPI, TSAPI), a wide variety of third-party providers evolved into what is today commonly referred to as the computer telephony integration (CTI) industry.

PACKET SWITCHES
Packet switching alters the time-sharing paradigm into a non-deterministic model. Session information is still transmitted as “groups” of digital bits, but these groups — now called packets — are no longer identifiable to a session or a terminal by their location in time or space. A communication session requires instead that every packet be provided with header information identifying (at a minimum) its origin and its destination. Each member terminal has to be always connected to the transmission medium in order to participate, and each has to read every header transmitted in order to detect and read its own packets.

The benefits to data communications include the ability to make packet length a matter of choice, and having packet rate become a function of processing capacity, bandwidth of the delivery medium, and competition from other sessions. Included in the performance tradeoffs for these benefits are irregular packet delays, varying throughput according to the number of active sessions, payload reduction due to packet overhead and error correction, and the absence of a centralized facility to measure, manage, or administer the system.

Adding Voice To Packet
The performance tradeoffs that worked so much to the advantage of machine-to-machine (data) communications made packet switching unusable for (person-to-person) real-time voice communications. Major developments that helped change this, ultimately leading to the feasibility of voice over packet, were in the fields of voice compression and high-speed (bandwidth) networking. Compression has an equivalent effect to increasing bandwidth by reducing the number of network bits required for a digital voice signal. When the effective path bandwidth gets to be sufficiently high, the ability to emulate TDM channels by artificially time spacing voice packets becomes a practicality (one of the premises of ATM). Even in the absence of TDM emulation by the delivery network, packets can be buffered at the terminal and played out at the desired constant packet rate provided the delivery rate stays sufficiently high.

The early commercial applications for packet-switched voice were targeted at the simple addition of voice to an existing data network, typically the Internet. The three approaches commonly used for this task added no (common control) gatekeeper functions. They can be most simply described as PBX-centric, gateway-centric, and remote access server (RAS)-centric configurations.

PBX-Centric Networking Model
The PBX-centric version relies on each PBX to act as a gatekeeper. Each has to provide network access, to select the far end gateway from which the dialed connection is to be completed, and to provide access control and call records. Limitations associated with this network model include the following:

• Minimum gateway port traffic values. Each port has a fixed association with a target gateway, and is not available for access to other gateways.
• Increasing the numbers of gateways will geometrically increase the number of ports required per gateway.
• Addition of a gateway also requires addition of a PBX for access purposes.
• Separate routing tables are required for each PBX. Routing table complexity increases geometrically with the addition of gateways.
• The originating PBX has no knowledge of the condition of target gateways, or of the IP network, and is therefore unable to accommodate fail-over or alternate routing.
• PC-to-telephone calls require the originating PC to know the IP number of the correct target gateway. The absence of an originating PBX requires that the terminating PBX have access control and the ability to provide billing information for PC-to-telephone calls.

Gateway-Centric Networking Model
In this model, each gateway assumes the gatekeeper role in providing the network access functions. These, at a minimum, would include interactive voice response (IVR), dialed number translation, event recording, and a programming interface.

Access can be provided directly from the PSTN, as well as from a PBX, and a PBX does not have to be added when adding another site. The site-specific allocation of circuits no longer applies, allowing any incoming circuit to be connected to any remote gateway based on the PSTN number dialed. The traffic value of gateway ports is therefore maximized, and adding more network sites no longer in itself requires the addition of ports to the existing sites.

It is still necessary with this configuration to maintain routing tables in each of the individual nodes, and each has to be provided with knowledge of all other nodes. No gateway has knowledge of the availability or busy status of any other, so that alternate routing, fail-over, or modified distribution functions are still unavailable. Traffic analyses and billing information cannot be provided in real time because call records that are stored locally first have be collected into a centralized database and consolidated in order to make them useful.

RAS-Centric Model
The RAS model is the same as the gateway-centric model, except that voice ports have been added to the data network access nodes as opposed to being provided in stand-alone gateways. In this case, the gatekeeper functions of access control and call recording are typically relegated to the connecting PBXs.

As with the two previously described configurations, once the session enters the data network through its originating port, the only thing available from the WAN is transport to its designated termination. There is no provision for the creation or application of value-added services by a delivery system created to provide point-to-point connections between the originating and terminating ports of two compatible RAS nodes on the same network.

Network-Centric Model
In the absence of a common control system, the configurations described above are essentially packet switching’s evolutionary counterparts to the step-by-step switches of fifty years ago. The addition of a stored program common control to a packet-switching network introduces the potential for creating, managing, administering, and measuring the many functions, features, and services that have come to be requirements in current voice and evolving multimedia telecommunications systems.

Packet switches provide an environment that is designed to accommodate distributed server-based processing wherein all control and payload communications share a common transport medium. A common control that has been designed to operate in this environment will by definition have to function as a coherent system component. Its capabilities become a function of its own design and structure, and its interfaces are definable in terms of the protocols it uses to communicate with its transport medium, with member terminations, with CTI applications processors, and with foreign systems. Though still a work in progress, the most widely accepted standard set of protocols for insuring a working level of interoperability among multi-vendor components is H.323. Like its circuit-switch counterpart EIA/TIA -464-A (PBX switching equipment for voiceband application), H.323 serves as an umbrella standard intended to provide rules for insuring a level of compatibility among components, and not as a blueprint for the creation of the components themselves.

CONCLUSION
The common control for a packet-based multimedia switching system is in actuality a self-contained, highly structured communications system. At one end of the functional spectrum, a common control implementation might be as simple as a single tasking gatekeeper capable of communicating a number translation using one protocol set such as H.323. At the other end of the spectrum, the implementation might be a multitasking system that includes multiple gatekeepers, provides multiple services, supports programmable user features, and interfaces with various member terminals, servers, and connected networks that use different industry standard protocols such as H.323, MGCP, and SIP.

Because such a common control is intended to be the central component of an integrated telecommunications switching system, its design capabilities will have the same kind of bearing on system size, expandability, flexibility, reliability, and manageability as does the common control of a legacy circuit-switching system. It is therefore appropriate to subject it to the same kinds of consideration that would apply to its mission-critical counterpart, the stored program common control of a circuit-switched voice system.

In the process of incorporating the growing requirements for multimedia (and voice in particular), packet switching technology has been going through some profound changes. These changes have a significant impact on all aspects of telecommunications design, application, and operation, and present a new set of responsibilities to designers, integrators, and end users alike. As a result of this convergence, the next major challenge will most likely be in the creation of a converged interconnect/integrator/VAR distribution infrastructure capable of helping this new technology find its promised land without encountering too many detours or stepping on too many land mines.

Harvey Kaufman is executive vice president of Netspeak Corporation. NetSpeak is a leading developer of advanced intelligent network (AIN) technologies for Internet telephony. NetSpeak’s comprehensive suite of IP telephony solutions enable service providers to increase revenue through enhanced value-added services, and enterprises to add low-cost communications capabilities to their existing infrastructure. For more information, visit the company’s Web site at www.netspeak.com.