- Email: [email protected]
ISDN in the year 2000
Comment ISDN in the year 2000
Kenneth L. Phillips
Historically,
Telecommunications plays a central role in fundamentally altering our relation to information and the manner in which we place value upon it. This article looks ahead to the state of telecommunications in the year 2000 and describes how technological advances are becoming obsolete with increasing speed. It discusses current technological advances which are nevertheless forming the backbone of planning ISDN, and comments on how ISDN is never likely to meet the objectives towards which it is directed unless basic planning assumptions are rethought in the light of the advances moving forward at unexpectedly rapid rates.
Kenneth L. Phillips is Chairman of the Committee of Corporate Telecom Users, and Vice President for Telecommunications Policy at Citicorp, 399 Park Avenue, 16th Floor, New York, USA (Tel: 212-559 4900). This paper was first presented at a meeting on Applications of ISDN: An International Perspective on User Research organized by Mijnchner Kreis and held on 24 June 1987 at the European Patent Office, Munich, West Germany.
TELECOMMUNICATIONS
POLICY
users of telecommunica-
and purveyors of information whether using smoke signals, semaphore systems, telegraph, telephone, point-to-point microwave, geostationary satellites, or fibre-based information transport networks, have all assumed a neutral relationship between the communications channel and message content. Such times, like those when milk was delivered each morning in shining glass bottles, are over. Advances in telecommunications have drastically and permanently altered our perception of the relationship between meaning, value and information. Beginning in the late 1970s through the 198Os, and continuing on through most of the 1990s the effects of advances in telecommunications on both our personal and professional lives happen alternatively as a result of progress in switching (processing) and transmission (transport). Imagine a model composed of a three-dimensional spiral if you will, moving upward in time, in which one side represents switching, the other transmission, and the distance from the central axis to any point on the spiral, a function of the rate of change. Though the history of telecommunications technologies is one of the most interesting and underestimated influences on our modern culture, our task here is to look to the future and describe the point on this spiral which will intersect the calendar year 2000 and beyond. Though intuition is often one’s most powerful tool in looking at technology tions
systems
December
1988
futures in a field where change is so rapid, there are some footprints in the sand. Ten years ago these were enhancements to digital switching such as pulse code modulation, and wave division multiplexing. To anticipate the future, we must appreciate our connectivity to the past. Today’s footprints are in the shape of superconductivity, multiple parallel processing, and very fast packet switching, to name a few. Of what do advances in these fields bode? As information flows through increasingly advanced systems and media of transport, the dynamic most accurately portraying the nature of the relationship among value, meaning and information increasingly approximates the inherent characteristics of the medium itself. For example, examine the effects of the increase in the velocity of information transfer made possible by photonics on financial information. Geopolitical boundaries are no longer meaningful in the various information markets. Product differentiation takes place as a function of first integrating independent streams of information, and then getting it into the hands of the appropriate user at an affordable cost, before the value of the information has decayed as a result of these same technologies falling into the hands of potential competitors. Winning economically at this game is not easy. A new dynamic has entered the picture: information half-life, the amount of time for a given pattern of information to lose 50% of the value of its meaning. The information, now converted
311
Comment
‘Telecommunications in the year 2000 will bear little resemblance to today’s service offerings’
312
to photonic forms of energy, begins to take on the characteristics of the medium: subatomic particles. Other ‘particle dynamics’ - entropy - also apply to information - its multiple existence at several locations simultaneously. As we move up the technology spiral, this ‘information halflife’ concatenates. Ten years ago a quotation of the price of a rare metal traded a few hours before on a market thousands of miles away might have been valuable. Today, it is as worthless as last year’s bus schedule. Telecommunications in the year 2000 will bear little resemblance to today’s service offerings, impressive though they are. Photonics will take on an expanded role at the fundamental processing level with lightbased logic gates comprising the heart of nodes of local processors linked in arrays forming hypercubes, replacing today’s linear, bus-based computer architecture. While suitable for a smaller set of applications than its serial or Turing precursor, the cubic, or hypercube architecture interconnecting multiple processors, each with local high-speed RAM, is ideal for telecommunications applications. Hypercubic processor architecture has advantages over the serial bus of infinitely greater expandability, enhanced survivability, obvious redundancy, and independent multi-tasking. In addition, cubic architecture, multiple processor computers have fewer processors per link than bus-based configurations, but will operate at staggeringly higher speeds. Intel was the first to develop a cube-based machine: the iPSC-VX series, which maxes out at 64 processors achieving 424 million floating point operations (MFLOPS) per second. Ironically, the greatest inherent limitation to be overcome in the effort to develop cubic processor-based capabilities for telecommunications applications is telecommunications itself. Traffic control over multiple processor buses in cubic arrays is increasingly non-trivial as speeds as high as 7000 MIPS can now be reached. Given the recently reported Bell Labs results of switching coherent light beams at .008 picoseconds, computers with heretofore unfathomable speed can be envis-
aged within the magnitude of our year 2000 spiral. While neurons in the human brain function at a frequency of only a few cycles per second, the complexity of our synapses makes the most elaborate cubic processing array look like child’s play. While the speed of optical computers promises to compensate for the relatively simplistic processor architecture, we are far from fulfilling the myth of the Golem: truly replicating human communications and information processing. Natural language processing. which has been the ‘black hole’ of R&D budgets for years, may now become reducible to workable problem sets leading to speech recognition, parsing, and generation running at acceptable rates for end users. This, in turn, will lead to the first forms of machine authorship. Users will access terminal devices that are ironically far simpler than today’s, and simply indicate with whom they wish to communicate, the type of information required, or the nature of the transaction to be completed, while basic applications continue to run in ‘background’ mode, just as our autonomic nervous system keeps watch over circulation, respiration and a host of other variables. Local cubic arrays of coprocessors in our homes and offices will access function-specific subnets via fibrebased fast packet switched channels (the day of the hierarchical network is over, too) and present the requested data based upon heuristic software resident in an ‘executive function’ processor, within the local cubic processor array. The ‘2B+D’ format at the base of today’s planning for ISDN’s standard is perhaps the first of many steps in this direction, but one likely to become obsolete even before it is implemented. The cubically architectured local nodes sending and receiving fast packets over fibre loops in ISDN-like formats may serve as a steersman for multiple expert systems providing medical advice in advnnce of likely illness, execution of financial transactions proactively, travel and entertainment options before reservations limits are exceeded, all forms of education, as well as basic voice and data telecommunications. In a word,
TELECOMMUNICATIONS
POLICY
December
1988
Commrnr
‘The sun may never come upon the dawn of the “Information Age” ’
TELECOMMUNICATIONS
POLICY
telecommunications should become infinitely more fransprrrenf as the metaphors bridging technologies become more natural. They may not, however; the sun may never come up on the dawn of the much touted ‘Information Age’ unless some rather radical realizations take place, and soon. I would like to suggest that the evolutionary, iterative approach being taken in different parts of the world today towards reaching agreement on an ISDN standard is not likely to result in a solution set having a half-life any longer, ironitally than the length of the deliberations in which such concurrence is being sought. Because the time lag associated with such international efforts is so very long, and because technology marches to its own beat. and not that of either regulators or policymakers, a much more revolutionary approach is required than we have seen thus far. In particular, we need to look at least towards an integrated services packet network; that it will be digital, goes without saying. As Jonathan Turner pointed out in the November 1986 issue of Selected Areas in Teleconmunications, there are several built-in problems with ISDN. Central to the issue is that current conceptualizations all anticipate a network model based around circuit switched voice, and a mixture of circuit-switched and packet switched data. This is patently obvious if one looks at the plan to provide two circuit switched 64 kbps channels plus a 16 kbps packet switched channel, (‘2B+D’). This might seem impressive to network planners who live in nations where mechanical switching remains the rule; however, the Japanese and North American experience should surely point up the fallacy in this sort of capacity planning. In such a configuration, channel capacity is something which cannot easily be altered to accommodate increasingly sophisticated applications. Given such a standard, how would an 80 kbps application be accommodated? What becomes immediately clear to any technologist who looks at the ISDN effort from the outside, is that an implicit goal of those doing the
December
1988
planning has been to maximize the protection of investment of capital in existing plant. Clearly, while a laudable goal given the tax structure some networks have to function in. such an approach is doomed to technological failure because of its inherent dualism: clearly two networks will be required wherever ISDN is implemented as it is presently envisaged. The only thing ‘integrated’ about ISDN is the name. Packet switching and circuit switching cannot be integrated, as they require vastly different technologies, have totally different economic trade-offs, and handle different types of traffic with varying degrees of success. To remedy the dualism. circuit switching must be abandoned altogether, in favour of novel approaches to networking based around fast packet switching. A homogeneous packet switching network holds many advantages over the presently conceptualized approach: Adaptive bandwidrh. Circuit switching systems cannot adapt to increases in the demand for bandwidth as new . apphcations are brought onto the network, while packetized networks supply the user with only the bandwidth actually needed. Unitary architecture. Whereas currently planned approaches to ISDN actually contemplate different switching networks for different types of information, a homogeneous packet switched network does not, and could pass along the savings associated with much lower common equipment costs. Traffic engineering efficiency. While most telecommunications engineers know that the majority of interactive data communications environments are characterized by spiked traffic patterns, few realize that voice is also not regular, since the sampling unit in the past has been an entire conversation. Look within a voice telephone conversation and one will quickly see, as the originators of TASI and other forms of analogue multiplexing did, that less than 44% of the actual bandwidth is used in a typical voice conversation. Under present packet switching techniques, even the im-
313
Commenr
pressive TASI ratio of [2x-l] significantly improved upon.
can be
general 0
‘The only thing integrated about ISDN is the name’
314
A brief diversion is called for here, as one may ask the obvious question: ‘If this efficiency argument is so true, then how come packet switching hasn’t already been employed for voice?’ History tells the story. Originally, of course, switching took place at the hands of telephone operators plugging cords into jacks. Strowger, an undertaker in a small city, wishing to eliminate the manual telephone operator who shunted all new business to his only competitor in town (her brother), invented a rachet-based electromechanical ‘step-by-step’ switch, which eventually gave way to larger panel offices. The revertive pulsing signalling system used in these ‘panel machines’ later played a significant role in Second World War cryptography efforts. These were in turn replaced by cross-bar switching, and eventually by the microprocessor controlled switching machines of today which can accommodate over 100 000 lines per switch as an end office, or 50 000 simultaneous conversations as a tandem. While unit costs per line of these large switches can be as low as $100-200, this is in large measure due to the economies of scale associated with the North American Network which today serves over 100 million telephones through over 10 000 local switching machines, though over 40% of the end office plant remains electromechanical. Not since the beginning of packet switching, with the ARPANET, has this technology played a dominant role in data communications, just as circuit switching has in voice or analogue telecommunications. In ARPA, the nodes are often very large timesharing processors called ‘hosts’; however, in recent years dedicated minicomputers have often replaced such hosts, frequently having specialized front-end processors to reduce the demand on the real-time processor hosts. ARPANET uses both relatively slow transmission links with modems operating at from 1 to 10 kbps, as well as end-to-end-digital facilities running at 56 kbps and 64 kbps. In looking at prospects for a truly integrated packet network, several
0
0
properties
can be identified:
Large scale operation equivalent to the public switched network (PSN) is required. Speed is very important because of the voice-based traffic, which typically requires end-to-end delays of less than 200ms. Baseline costs must be low; voice users in particular will not pay for a better system if it costs more.
Where can infrastructure capable of meeting these goals be found? A natural match, it must be recognized, is the existing PSN, which is based around switching systems which already support 1.54 Mbps interfaces carrying 24 voice calls in the 63 kbps format. Traditional packet switching networks typically interconnect at much narrower bandwidths, often requiring modems to contend with analogue transmission channels, and incur far higher bit-error rates. A perfectly conceivable advantage of utilizing the PSN for ISPN is the amount of specialpurpose equipment already in place in connection with the switching function. Conventional packet switching systems often use a general-purpose computer which must literally do some processing of every packet, leading to a far higher likelihood of throughput issues. Given the generic planning directives just mentioned, it is possible to begin to envisage what an ISPN might look like. From an operational viewpoint, the network must be fast, especially because of carrying voice traffic. (looms-150ms is generally an acceptable delay; more than 400ms-500ms is distracting and confusing to the human ear.) Bandwidth capacity must be large. Realistically speaking, the bandwidth capacity of switched channels should be considerably broader than those available over today’s PSN. This increased bandwidth and speed cannot however be at a higher unit cost, as work undertaken by the Committee of Corporate Telecommunications Users in the USA has shown. There is a definite limit to users’ fascination with the incremental cost of advanced technology. As Turner points out, many lessons may be learned from our design ex-
TELECOMMUNICATIONS
POLICY
December
1988
Commrnr
‘Many lessons may be learned from our design experience in providing POTS’
TELECOMMUNICATIONS
POLICY
perience in providing POTS, since the domestic network utilizes transmission facilities having bandwidths commonly up to 1.5 Mbps, each multiplexed into a 2464 kbps sub-channel format. In the past, packet switching systems had to be fed off much narrower channels from modems, resulting in higher bit error rates. Another lesson from the public switched network which is often neglected is that while large packet nets operated by nondominant carriers typically use general-purpose computers to both drive the network and handle traffic and flow control, larger nationwide carrier-operated networks have costjustified and implemented specialpurpose processing nodes. The result is a considerable difference in throughput efficiency. Given the growth in the traffic handling ability augured by the advances in technology already discussed, the day may arrive sooner than expected when the use of general-purpose devices is no longer feasible in packet nets because of having to actually process each and every packet in the bit stream. Given an end-to-end digital environment, new packet protocols could be developed, especially at the lower end of the IS0 model. Link layer error correction could theoretically be devised and provided on an end-to-end basis. This could also open new competitive business opportunities, though the possibility of proliferating multiple, incompatible end-toend error correction and flow control methods would constitute the ‘down side’ risk. A T-l, 1.54 Mbps link functioning in a circuit switched environment generally supports 24 simultaneous voice connections. In an end-to-end digital transmission environment approximately 50 simultaneous conversations could be accommodated. Somewhat over 100 conversations may be carried by a 1.5 Mbps channel using adaptive differential pulse code modulation (ADPCM). At this density a 32 Mbps format would mean that 50 000 simultaneous conversations could be carried by 1200 full duplex channels. In terms of net nodal throughput this means capacity of over 2 Mbps.
December
1988
Preliminary network architecture design is now possible. Packet switches (PS) will be assumed to be able to accommodate up to 1000 high speed channels (HSCs) using ADPCM at the 32 kbps density. Direct, high-speed access to the network would be available to business/commercial users, users while consumer/residential would tie in over medium speed channels (MSCs) operating at 100 kbps. Customer premises equipment (CPE) supported by the network would take many forms, depending on the application being supported. Single line’ applications, telephone set-like obviously comprising the low end of the offerings, could be based around a simple eight-bit microprocessor. Obviously, multi-line interfaces or those used to support simultaneous voice/data (SVD) applications would require more complex local control functions, but happily, from a network standpoint, merely affects the consumption of incremental synchronous cycles. Network interface timing nodes (NINs) interconnect to either external large-scale nets, or on the other side, LANs. NINs provide network interconnectivity, protection, cost accounting data collecting, and multiple small customer interconnection capabilities. A possible configuration supporting end users directly, or customers in a carrier environment, might consist of 125 NINs accessing a PS each via an HSC, thus supporting approximately 60 000 subscribers per network node in combinations of single subscriber controllers, multiline controllers, up through LAN, PBX, and full baseband interfaces for special applications. An issue of great importance will be the development of a migration strategy from today’s circuit switched facilities to the ISPN world. NINs will have to be designed, initially at least, with interfaces to today’s analogue DC loops and MF/DTMF signalling systems. Obviously, the efficiency of the NINs would be reduced, probably to 2000 customers per mode during the migration period. A planning matter which has already come up in the provisioning of point-to-point services of all kinds over end-to-end digital channels, is
315
that of ‘virtual circuits’. In the ISPN world, a request for a ‘dedicated’ link would be accepted or rejected by the network controllers depending upon real-time demand characteristics and quite possibly class markers. Dedicated voice would of course ‘look’ no different to the network than dedicated data, simply a potentially narrower channel of 12 kbps. Greater networking efficiencies would be possible in the virtual circuit environment supporting point-to-point applications since if bandwidth is asymmetrically less (or greater) in one direction, potentially less excess transmission capacity would be reallocated. One new characteristic which the ISPN will have to accommodate is that of individually addressed packets which are not part of either an existing connection or a repeated control sequence. Hence, link, network, and terminal level protocols will have to embody control functions capable of detecting the presence of such packets or datagrams, monitor and pass status, and deal appropriately with as yet undelivered packets of datagrams. A number of potential design issues shall have to be addressed: 0
‘We must not be hypnotized by blinking boxes’
0 0
0 0 0
Customer/user applicationsdependent protocol processing. Basic service/voice protocols. Connection protocols. Timing/sequencing protocols. Internetworking handshaking protocols. Datagramming.
Additional features may have to be added to handle special formats at the packet level, ie, variable length/field packets, and overhead information packets which typically could reach either 144 or 256 bytes. Conclusion The purpose of this discussion has been five-fold. First, to focus attention on the much overlooked fact of the central role that telecommunications has played, and will continue to play in fundamentally altering our relation to information and the manner in which we place value upon it. Secondly, to point to a few developments such as cubic array processors and backplane superconductivity, which
316
lurk just over the event horizon and which will render much of the planning activities going on today obsolete before even being implemented, because of the error inherent in assuming a linear relationship among the velocity of technological changes in the past, regulation, and those coming in the future. Thirdly, as an example of an early victim of this error, the ISDN effort is cited. Though a valuable and well meaning international effort, it is likely never to meet the objectives towards which it is directed, unless basic planning assumptions are rethought in light of the advances moving forward at unexpectedly rapid rates. Fourth, a new network design is proposed based around a form of fast packet switching more fully able to exploit the emerging changes in technology brought on by photonic logic gates, superconducting buses, and cubically arrayed multi-processors. Jonathan S. Turner’s proposals in this area appear to be especially ‘on target’. Fifth, and finally, we must not be hypnotized by blinking boxes and other religious icons comprising today’s Zeitgeist, for as we speak, governments exert protectionist restrictions on the free flow of information across geopolitical boundaries, threatening to return us to an age of electronic feudalism. Such restrictions will become even more of a legal fiction in the years ahead. As economics become increasingly based upon the flow of photons representing ‘value-added’ meaning, and therefore capital, political entities which today remain so closed so as to feel the necessity to guard and license photocopy machines, eavesdrop on international telecommunications gateways, or outright remain cut off from global networks, will eventually be confronted with a ‘do or die’ choice. Yet the position of many delegations to the upcoming World Administrative Telegraph and Telephone Conference is essentially to increase the grasp of regulation through modifying the working definitions of services to include ‘enhanced’ information and service providers. To think that attempting to control the free flow of in-
TELECOMMUNICATIONS
POLICY
December
1988
Comment
formation, and therefore capital, benefits any political entity in the long run, or the EC as a whole, is to ignore totally the lessons of history. Telecommunications is rapidly becoming the most efficient vehicle for the redistribution of capital. As perhaps as unrealistically altruistic as it may sound today, technologies of
TELECOMMUNICATIONS
POLICY
December
1988
telecommunications, with their capability to link anyone with anyone at a unit cost inversely proportional to distance, if allowed to blossom absent medieval regulation and protectionist government interference, can become the most potent social and political force moving in the direction of world peace.
317
Kenneth L. Phillips
Historically,
Telecommunications plays a central role in fundamentally altering our relation to information and the manner in which we place value upon it. This article looks ahead to the state of telecommunications in the year 2000 and describes how technological advances are becoming obsolete with increasing speed. It discusses current technological advances which are nevertheless forming the backbone of planning ISDN, and comments on how ISDN is never likely to meet the objectives towards which it is directed unless basic planning assumptions are rethought in the light of the advances moving forward at unexpectedly rapid rates.
Kenneth L. Phillips is Chairman of the Committee of Corporate Telecom Users, and Vice President for Telecommunications Policy at Citicorp, 399 Park Avenue, 16th Floor, New York, USA (Tel: 212-559 4900). This paper was first presented at a meeting on Applications of ISDN: An International Perspective on User Research organized by Mijnchner Kreis and held on 24 June 1987 at the European Patent Office, Munich, West Germany.
TELECOMMUNICATIONS
POLICY
users of telecommunica-
and purveyors of information whether using smoke signals, semaphore systems, telegraph, telephone, point-to-point microwave, geostationary satellites, or fibre-based information transport networks, have all assumed a neutral relationship between the communications channel and message content. Such times, like those when milk was delivered each morning in shining glass bottles, are over. Advances in telecommunications have drastically and permanently altered our perception of the relationship between meaning, value and information. Beginning in the late 1970s through the 198Os, and continuing on through most of the 1990s the effects of advances in telecommunications on both our personal and professional lives happen alternatively as a result of progress in switching (processing) and transmission (transport). Imagine a model composed of a three-dimensional spiral if you will, moving upward in time, in which one side represents switching, the other transmission, and the distance from the central axis to any point on the spiral, a function of the rate of change. Though the history of telecommunications technologies is one of the most interesting and underestimated influences on our modern culture, our task here is to look to the future and describe the point on this spiral which will intersect the calendar year 2000 and beyond. Though intuition is often one’s most powerful tool in looking at technology tions
systems
December
1988
futures in a field where change is so rapid, there are some footprints in the sand. Ten years ago these were enhancements to digital switching such as pulse code modulation, and wave division multiplexing. To anticipate the future, we must appreciate our connectivity to the past. Today’s footprints are in the shape of superconductivity, multiple parallel processing, and very fast packet switching, to name a few. Of what do advances in these fields bode? As information flows through increasingly advanced systems and media of transport, the dynamic most accurately portraying the nature of the relationship among value, meaning and information increasingly approximates the inherent characteristics of the medium itself. For example, examine the effects of the increase in the velocity of information transfer made possible by photonics on financial information. Geopolitical boundaries are no longer meaningful in the various information markets. Product differentiation takes place as a function of first integrating independent streams of information, and then getting it into the hands of the appropriate user at an affordable cost, before the value of the information has decayed as a result of these same technologies falling into the hands of potential competitors. Winning economically at this game is not easy. A new dynamic has entered the picture: information half-life, the amount of time for a given pattern of information to lose 50% of the value of its meaning. The information, now converted
311
Comment
‘Telecommunications in the year 2000 will bear little resemblance to today’s service offerings’
312
to photonic forms of energy, begins to take on the characteristics of the medium: subatomic particles. Other ‘particle dynamics’ - entropy - also apply to information - its multiple existence at several locations simultaneously. As we move up the technology spiral, this ‘information halflife’ concatenates. Ten years ago a quotation of the price of a rare metal traded a few hours before on a market thousands of miles away might have been valuable. Today, it is as worthless as last year’s bus schedule. Telecommunications in the year 2000 will bear little resemblance to today’s service offerings, impressive though they are. Photonics will take on an expanded role at the fundamental processing level with lightbased logic gates comprising the heart of nodes of local processors linked in arrays forming hypercubes, replacing today’s linear, bus-based computer architecture. While suitable for a smaller set of applications than its serial or Turing precursor, the cubic, or hypercube architecture interconnecting multiple processors, each with local high-speed RAM, is ideal for telecommunications applications. Hypercubic processor architecture has advantages over the serial bus of infinitely greater expandability, enhanced survivability, obvious redundancy, and independent multi-tasking. In addition, cubic architecture, multiple processor computers have fewer processors per link than bus-based configurations, but will operate at staggeringly higher speeds. Intel was the first to develop a cube-based machine: the iPSC-VX series, which maxes out at 64 processors achieving 424 million floating point operations (MFLOPS) per second. Ironically, the greatest inherent limitation to be overcome in the effort to develop cubic processor-based capabilities for telecommunications applications is telecommunications itself. Traffic control over multiple processor buses in cubic arrays is increasingly non-trivial as speeds as high as 7000 MIPS can now be reached. Given the recently reported Bell Labs results of switching coherent light beams at .008 picoseconds, computers with heretofore unfathomable speed can be envis-
aged within the magnitude of our year 2000 spiral. While neurons in the human brain function at a frequency of only a few cycles per second, the complexity of our synapses makes the most elaborate cubic processing array look like child’s play. While the speed of optical computers promises to compensate for the relatively simplistic processor architecture, we are far from fulfilling the myth of the Golem: truly replicating human communications and information processing. Natural language processing. which has been the ‘black hole’ of R&D budgets for years, may now become reducible to workable problem sets leading to speech recognition, parsing, and generation running at acceptable rates for end users. This, in turn, will lead to the first forms of machine authorship. Users will access terminal devices that are ironically far simpler than today’s, and simply indicate with whom they wish to communicate, the type of information required, or the nature of the transaction to be completed, while basic applications continue to run in ‘background’ mode, just as our autonomic nervous system keeps watch over circulation, respiration and a host of other variables. Local cubic arrays of coprocessors in our homes and offices will access function-specific subnets via fibrebased fast packet switched channels (the day of the hierarchical network is over, too) and present the requested data based upon heuristic software resident in an ‘executive function’ processor, within the local cubic processor array. The ‘2B+D’ format at the base of today’s planning for ISDN’s standard is perhaps the first of many steps in this direction, but one likely to become obsolete even before it is implemented. The cubically architectured local nodes sending and receiving fast packets over fibre loops in ISDN-like formats may serve as a steersman for multiple expert systems providing medical advice in advnnce of likely illness, execution of financial transactions proactively, travel and entertainment options before reservations limits are exceeded, all forms of education, as well as basic voice and data telecommunications. In a word,
TELECOMMUNICATIONS
POLICY
December
1988
Commrnr
‘The sun may never come upon the dawn of the “Information Age” ’
TELECOMMUNICATIONS
POLICY
telecommunications should become infinitely more fransprrrenf as the metaphors bridging technologies become more natural. They may not, however; the sun may never come up on the dawn of the much touted ‘Information Age’ unless some rather radical realizations take place, and soon. I would like to suggest that the evolutionary, iterative approach being taken in different parts of the world today towards reaching agreement on an ISDN standard is not likely to result in a solution set having a half-life any longer, ironitally than the length of the deliberations in which such concurrence is being sought. Because the time lag associated with such international efforts is so very long, and because technology marches to its own beat. and not that of either regulators or policymakers, a much more revolutionary approach is required than we have seen thus far. In particular, we need to look at least towards an integrated services packet network; that it will be digital, goes without saying. As Jonathan Turner pointed out in the November 1986 issue of Selected Areas in Teleconmunications, there are several built-in problems with ISDN. Central to the issue is that current conceptualizations all anticipate a network model based around circuit switched voice, and a mixture of circuit-switched and packet switched data. This is patently obvious if one looks at the plan to provide two circuit switched 64 kbps channels plus a 16 kbps packet switched channel, (‘2B+D’). This might seem impressive to network planners who live in nations where mechanical switching remains the rule; however, the Japanese and North American experience should surely point up the fallacy in this sort of capacity planning. In such a configuration, channel capacity is something which cannot easily be altered to accommodate increasingly sophisticated applications. Given such a standard, how would an 80 kbps application be accommodated? What becomes immediately clear to any technologist who looks at the ISDN effort from the outside, is that an implicit goal of those doing the
December
1988
planning has been to maximize the protection of investment of capital in existing plant. Clearly, while a laudable goal given the tax structure some networks have to function in. such an approach is doomed to technological failure because of its inherent dualism: clearly two networks will be required wherever ISDN is implemented as it is presently envisaged. The only thing ‘integrated’ about ISDN is the name. Packet switching and circuit switching cannot be integrated, as they require vastly different technologies, have totally different economic trade-offs, and handle different types of traffic with varying degrees of success. To remedy the dualism. circuit switching must be abandoned altogether, in favour of novel approaches to networking based around fast packet switching. A homogeneous packet switching network holds many advantages over the presently conceptualized approach: Adaptive bandwidrh. Circuit switching systems cannot adapt to increases in the demand for bandwidth as new . apphcations are brought onto the network, while packetized networks supply the user with only the bandwidth actually needed. Unitary architecture. Whereas currently planned approaches to ISDN actually contemplate different switching networks for different types of information, a homogeneous packet switched network does not, and could pass along the savings associated with much lower common equipment costs. Traffic engineering efficiency. While most telecommunications engineers know that the majority of interactive data communications environments are characterized by spiked traffic patterns, few realize that voice is also not regular, since the sampling unit in the past has been an entire conversation. Look within a voice telephone conversation and one will quickly see, as the originators of TASI and other forms of analogue multiplexing did, that less than 44% of the actual bandwidth is used in a typical voice conversation. Under present packet switching techniques, even the im-
313
Commenr
pressive TASI ratio of [2x-l] significantly improved upon.
can be
general 0
‘The only thing integrated about ISDN is the name’
314
A brief diversion is called for here, as one may ask the obvious question: ‘If this efficiency argument is so true, then how come packet switching hasn’t already been employed for voice?’ History tells the story. Originally, of course, switching took place at the hands of telephone operators plugging cords into jacks. Strowger, an undertaker in a small city, wishing to eliminate the manual telephone operator who shunted all new business to his only competitor in town (her brother), invented a rachet-based electromechanical ‘step-by-step’ switch, which eventually gave way to larger panel offices. The revertive pulsing signalling system used in these ‘panel machines’ later played a significant role in Second World War cryptography efforts. These were in turn replaced by cross-bar switching, and eventually by the microprocessor controlled switching machines of today which can accommodate over 100 000 lines per switch as an end office, or 50 000 simultaneous conversations as a tandem. While unit costs per line of these large switches can be as low as $100-200, this is in large measure due to the economies of scale associated with the North American Network which today serves over 100 million telephones through over 10 000 local switching machines, though over 40% of the end office plant remains electromechanical. Not since the beginning of packet switching, with the ARPANET, has this technology played a dominant role in data communications, just as circuit switching has in voice or analogue telecommunications. In ARPA, the nodes are often very large timesharing processors called ‘hosts’; however, in recent years dedicated minicomputers have often replaced such hosts, frequently having specialized front-end processors to reduce the demand on the real-time processor hosts. ARPANET uses both relatively slow transmission links with modems operating at from 1 to 10 kbps, as well as end-to-end-digital facilities running at 56 kbps and 64 kbps. In looking at prospects for a truly integrated packet network, several
0
0
properties
can be identified:
Large scale operation equivalent to the public switched network (PSN) is required. Speed is very important because of the voice-based traffic, which typically requires end-to-end delays of less than 200ms. Baseline costs must be low; voice users in particular will not pay for a better system if it costs more.
Where can infrastructure capable of meeting these goals be found? A natural match, it must be recognized, is the existing PSN, which is based around switching systems which already support 1.54 Mbps interfaces carrying 24 voice calls in the 63 kbps format. Traditional packet switching networks typically interconnect at much narrower bandwidths, often requiring modems to contend with analogue transmission channels, and incur far higher bit-error rates. A perfectly conceivable advantage of utilizing the PSN for ISPN is the amount of specialpurpose equipment already in place in connection with the switching function. Conventional packet switching systems often use a general-purpose computer which must literally do some processing of every packet, leading to a far higher likelihood of throughput issues. Given the generic planning directives just mentioned, it is possible to begin to envisage what an ISPN might look like. From an operational viewpoint, the network must be fast, especially because of carrying voice traffic. (looms-150ms is generally an acceptable delay; more than 400ms-500ms is distracting and confusing to the human ear.) Bandwidth capacity must be large. Realistically speaking, the bandwidth capacity of switched channels should be considerably broader than those available over today’s PSN. This increased bandwidth and speed cannot however be at a higher unit cost, as work undertaken by the Committee of Corporate Telecommunications Users in the USA has shown. There is a definite limit to users’ fascination with the incremental cost of advanced technology. As Turner points out, many lessons may be learned from our design ex-
TELECOMMUNICATIONS
POLICY
December
1988
Commrnr
‘Many lessons may be learned from our design experience in providing POTS’
TELECOMMUNICATIONS
POLICY
perience in providing POTS, since the domestic network utilizes transmission facilities having bandwidths commonly up to 1.5 Mbps, each multiplexed into a 2464 kbps sub-channel format. In the past, packet switching systems had to be fed off much narrower channels from modems, resulting in higher bit error rates. Another lesson from the public switched network which is often neglected is that while large packet nets operated by nondominant carriers typically use general-purpose computers to both drive the network and handle traffic and flow control, larger nationwide carrier-operated networks have costjustified and implemented specialpurpose processing nodes. The result is a considerable difference in throughput efficiency. Given the growth in the traffic handling ability augured by the advances in technology already discussed, the day may arrive sooner than expected when the use of general-purpose devices is no longer feasible in packet nets because of having to actually process each and every packet in the bit stream. Given an end-to-end digital environment, new packet protocols could be developed, especially at the lower end of the IS0 model. Link layer error correction could theoretically be devised and provided on an end-to-end basis. This could also open new competitive business opportunities, though the possibility of proliferating multiple, incompatible end-toend error correction and flow control methods would constitute the ‘down side’ risk. A T-l, 1.54 Mbps link functioning in a circuit switched environment generally supports 24 simultaneous voice connections. In an end-to-end digital transmission environment approximately 50 simultaneous conversations could be accommodated. Somewhat over 100 conversations may be carried by a 1.5 Mbps channel using adaptive differential pulse code modulation (ADPCM). At this density a 32 Mbps format would mean that 50 000 simultaneous conversations could be carried by 1200 full duplex channels. In terms of net nodal throughput this means capacity of over 2 Mbps.
December
1988
Preliminary network architecture design is now possible. Packet switches (PS) will be assumed to be able to accommodate up to 1000 high speed channels (HSCs) using ADPCM at the 32 kbps density. Direct, high-speed access to the network would be available to business/commercial users, users while consumer/residential would tie in over medium speed channels (MSCs) operating at 100 kbps. Customer premises equipment (CPE) supported by the network would take many forms, depending on the application being supported. Single line’ applications, telephone set-like obviously comprising the low end of the offerings, could be based around a simple eight-bit microprocessor. Obviously, multi-line interfaces or those used to support simultaneous voice/data (SVD) applications would require more complex local control functions, but happily, from a network standpoint, merely affects the consumption of incremental synchronous cycles. Network interface timing nodes (NINs) interconnect to either external large-scale nets, or on the other side, LANs. NINs provide network interconnectivity, protection, cost accounting data collecting, and multiple small customer interconnection capabilities. A possible configuration supporting end users directly, or customers in a carrier environment, might consist of 125 NINs accessing a PS each via an HSC, thus supporting approximately 60 000 subscribers per network node in combinations of single subscriber controllers, multiline controllers, up through LAN, PBX, and full baseband interfaces for special applications. An issue of great importance will be the development of a migration strategy from today’s circuit switched facilities to the ISPN world. NINs will have to be designed, initially at least, with interfaces to today’s analogue DC loops and MF/DTMF signalling systems. Obviously, the efficiency of the NINs would be reduced, probably to 2000 customers per mode during the migration period. A planning matter which has already come up in the provisioning of point-to-point services of all kinds over end-to-end digital channels, is
315
that of ‘virtual circuits’. In the ISPN world, a request for a ‘dedicated’ link would be accepted or rejected by the network controllers depending upon real-time demand characteristics and quite possibly class markers. Dedicated voice would of course ‘look’ no different to the network than dedicated data, simply a potentially narrower channel of 12 kbps. Greater networking efficiencies would be possible in the virtual circuit environment supporting point-to-point applications since if bandwidth is asymmetrically less (or greater) in one direction, potentially less excess transmission capacity would be reallocated. One new characteristic which the ISPN will have to accommodate is that of individually addressed packets which are not part of either an existing connection or a repeated control sequence. Hence, link, network, and terminal level protocols will have to embody control functions capable of detecting the presence of such packets or datagrams, monitor and pass status, and deal appropriately with as yet undelivered packets of datagrams. A number of potential design issues shall have to be addressed: 0
‘We must not be hypnotized by blinking boxes’
0 0
0 0 0
Customer/user applicationsdependent protocol processing. Basic service/voice protocols. Connection protocols. Timing/sequencing protocols. Internetworking handshaking protocols. Datagramming.
Additional features may have to be added to handle special formats at the packet level, ie, variable length/field packets, and overhead information packets which typically could reach either 144 or 256 bytes. Conclusion The purpose of this discussion has been five-fold. First, to focus attention on the much overlooked fact of the central role that telecommunications has played, and will continue to play in fundamentally altering our relation to information and the manner in which we place value upon it. Secondly, to point to a few developments such as cubic array processors and backplane superconductivity, which
316
lurk just over the event horizon and which will render much of the planning activities going on today obsolete before even being implemented, because of the error inherent in assuming a linear relationship among the velocity of technological changes in the past, regulation, and those coming in the future. Thirdly, as an example of an early victim of this error, the ISDN effort is cited. Though a valuable and well meaning international effort, it is likely never to meet the objectives towards which it is directed, unless basic planning assumptions are rethought in light of the advances moving forward at unexpectedly rapid rates. Fourth, a new network design is proposed based around a form of fast packet switching more fully able to exploit the emerging changes in technology brought on by photonic logic gates, superconducting buses, and cubically arrayed multi-processors. Jonathan S. Turner’s proposals in this area appear to be especially ‘on target’. Fifth, and finally, we must not be hypnotized by blinking boxes and other religious icons comprising today’s Zeitgeist, for as we speak, governments exert protectionist restrictions on the free flow of information across geopolitical boundaries, threatening to return us to an age of electronic feudalism. Such restrictions will become even more of a legal fiction in the years ahead. As economics become increasingly based upon the flow of photons representing ‘value-added’ meaning, and therefore capital, political entities which today remain so closed so as to feel the necessity to guard and license photocopy machines, eavesdrop on international telecommunications gateways, or outright remain cut off from global networks, will eventually be confronted with a ‘do or die’ choice. Yet the position of many delegations to the upcoming World Administrative Telegraph and Telephone Conference is essentially to increase the grasp of regulation through modifying the working definitions of services to include ‘enhanced’ information and service providers. To think that attempting to control the free flow of in-
TELECOMMUNICATIONS
POLICY
December
1988
Comment
formation, and therefore capital, benefits any political entity in the long run, or the EC as a whole, is to ignore totally the lessons of history. Telecommunications is rapidly becoming the most efficient vehicle for the redistribution of capital. As perhaps as unrealistically altruistic as it may sound today, technologies of
TELECOMMUNICATIONS
POLICY
December
1988
telecommunications, with their capability to link anyone with anyone at a unit cost inversely proportional to distance, if allowed to blossom absent medieval regulation and protectionist government interference, can become the most potent social and political force moving in the direction of world peace.
317