by FRANK BACKMAN
IBM Design Center Gaithersburg, Maryland
ABSTRACT
Once the furor over the legal and regulatory issues of electronic funds transfer dies down, the data process-ing industry will find several complications in the development of a systems architecture suitable for implementing a nationwide system. Apart from inter-face difficulties presented by the equipment and soft-ware designs of competing manufacturers, there are complications to the control of such a system that appear to merit the attention of the data processing industry. The purpose of this paper is to stimulate the exchange of opinions on how the industry should cope with differences in end-user protocols, different routing techniques, different methods of controlling terminal devices, the value of various communication line dis-ciplines, and the overall management of a nationwide network.
BACKGROUND
For the past several years the move toward an electronic funds transfer system has been gaining mo-mentum. The Federal Reserve System already trans-fers vast sums of money electrically; banking termi-nals are popping up in supermarkets and airports; the battle lines between the thrift institutions and the commercial banks have been drawn; and reliable digi-tal communications are becoming available at an un-precedented rate. Even the most strident slow-payers can be accommodated in the systems being proposed.
It seems inevitable that the flood of paper that is engulfing the financial industry is about to be abated by a worldwide network of digital computers and ter-minals. After all, stored-program digital devices can do anything!
The problem is that they don't; and there seems to be a labyrinth of data processing issues that will become critical, once the furor over legal and regula-tory rules dies dOWR
To address these data processing issues, the Federal Home Loan Bank Board, in March 1975, invited in-dustry to suggest an electronic funds transfer system that could evolve from the regional needs of a mythical
147
metropolitan area called Middletown. The Federal Home Loan Bank Board provided a scenario that illus-trated some of the problems being faced by the archi-tects of an electronic funds transfer system today. The scenario carried the evolution of the EFTS system through several stages in which Savings and Loan Associations entered agreements with local merchants to handle debit and credit card transactions, installed banking terminals at shopping centers, added new functions, and included new S & L's into the evolving system. The scenario, which included enterprises with such picturesque names as Korn Krib Supermarkets and Marshall Prairie Department Stores, was made even more vivid by the technical difficulties posed at each stage of the evolution.
Several computer and terminal manufacturers' equipment was involved, and an equally complicated mixture of transaction types, communication line dis-ciplines and business practices were represented. Re-spondents to the FHLBB invitation were not requested to submit the detailed design of a system that would track the evolution of the Middletown scenario through its various stages, but rather to postulate a system that would solve similar problems scaled for a metropolitan area with a population of from two million to five million people.
The purpose of this paper is to identify some of these problems that are of a data processing nature (ignor-ing legal and regulatory issues), to suggest promis(ignor-ing directions, and to stimulate discussion among the archi-tects of the future Electronic Funds Transfer System.
.LA;LSSUMPTIONS
The arrangement of the institutions on the topo-graphic maps at the end of this paper has been syn-thesized from the demographic information in the scenario developed by the Federal Home Loan Bank Board and illustrates several points, other than purely technical considerations, that will affect the evolution of an electronic funds transfer switch. The evolution shown for the Middletown environment is intended to represent only a plausible evolution through the vari-ous stages under the assumption that no consortium is
formed to bypass stages. In actual practice, the de-cision to install a regional switch may be reached at any time, so that some of the technical problems in the expanded scenario need never be addressed.
In the Middletown environment, we have assumed that some transactions may cross a state line. Al-though this consideration is difficult to cope with in the design of a generalized system, it presents a real-world problem because of the variety of inter- and intra-state communication line tariffs and because of state banking laws. This is a non-technical considera-tion that will influence the topology of the network.
The location of the thrift institution servicers in the downtown location suggests the possibility of colloca-tion which would, if implemented, have a profound impact on the design of the communication system. We have, however, assumed that the thrift institution servicers are physically separated from one another and that communications lines are required.
The geographic layout of Middletown (as shown in stage 1 and 2a), suggests that these stages would be implemented by a combination of point-to-point and multipoint lines rather than by point-to-point lines only as was illustrated in the original scenario. In addition, reliability considerations may cause First Federal to install two separate multipoint lines in each of its branches. This would permit each branch to tie half its terminals to each line so that business could continue normally in the event of a single line failure.
The practicality of doing this is dictated by the prices and the line routing of the local common carriers and is another illustration of a non-technical consideration that wiII influence network topology.
In the later stages of the evolution, we have assumed that the network would evolve as in the scenario with the thrift institutions and merchants entering agree-ments with one another at each stage. In order to pre-serve the staging of the scenario we have also assumed that the rival thrift institution servicers would some-how be able to overcome the technical problems of
"looking like one another's terminals" and would accommodate one another's customers by developing and installing the software necessary to cope with the different transaction formats used by the various thrift institutions. This latter assumption, of course, is opti-mistic, but illustrates a natural tendency toward the evolution of a hybrid network. We have however, assumed that at some point (illustrated in stage 3D) a regional switch is installed and that S & L's sub-scribe to it through their servicers.
The remainder of this paper describes some key elements of network design and suggests that, once non-technical problems have been overcome and ade-quate computational "horsepower" has been planned for, the network design is dominated by data process-ing software and procedural considerations rather than by communications considerations.
NETWORK DESIGN CONSIDERATIONS
Figure 1 shows a simple hierarchical structure for an EFTS network In practice, the network will be more complex than that shown, since it will frequently be advantageous for terminals to access the switch directly rather than via the computers serving the thrift institutions. (In the scenario, for example, Sec-ond Federal may decide to attach their teller terminals and their Merchant Operated Terminals at Fast Food directly to the switch, rather than through the FHLB data center 250 miles away. This decision would save 250 miles of communication line at the expense of increased switching load that would be generated by any "on us" transactions. This situation, can be ex-pected in the actual EFTS network. (The decision will also be influenced by the compatibility of Second Fed-eral's terminals with the switch.) Furthermore, traffic considerations will undoubtedly justify additional in-tercommunication links among terminals. N everthe-less, even the tree structure of Figure 1 illustrates five key considerations that will dominate the design of the EFTS network. These five considerations, listed on Figure 1 form the basis for IBM's Systems Network Architecture (SNA), and are being addressed by other implementers of specialized computer networks. As yet, however, there is no set of widely accepted dards for computer networking. No matter what stan-dards are adopted by the Federal Home Loan Banking System, these management problems must be addressed
1 NETWORK CONTROL 2 PATH CONTROL (ROUTING) 3 LINK CONTROL
TERMINALS
00
4 DEVICE CONTROL iNETWORK SECURITvl 5 END USER PROTOCOES ICUSTOMER SECURITYI
Figure l-A simple topography
FUTURE CONNECTIONS TO OTHER REGIONAL SWITCHES
... L\.re Computers Read:"T for the Checkless 149
in the design of the EFTS network and the participat-ing thrift institution servicers must modify their sys-tems to conform to the EFTS network rules.
1. Network Control--In order to allow the future interconnection of EFTS switches, some form of distributed control must be used. Control includes responsibility for the equipment configuration in the network subordinate to the control locations, the loading of the communication programs in the processors and cluster controllers of the thrift in-stitution servicers, the reaction to outages, and the activation and deactivation of equipment in re-sponse to changing workloads. The Systems N et-work Architecture, as currently implemented by IBM, concentrates control at the highest level of a tree-structured network and distributes responsi-bility for its execution to subordinate elements. In the EFTS, control would reside at the switch with the bulk of the responsibility delegated to the thrift institution servicers.
Special arrangements must be made to crosstell status information between thrift institution ser-vicers when direct links are installed. (In the scenario, for example, it is conceivable that a direct link could be established between two of the data centers, bypassing the switch, and dedicated to high volume transactions between these servicers. It is important that the overall network design not pre-clude this sort of connection if the member banks feel that it is to their advantage. This situation is invisible to the switch but requires the development of special protocols by the servicers. Further com-plications to network control are discussed later and must be resolved by the standards adopted by the Federal Home Loan Banking System.
2. Path Control, or routing, is the second key con-sideration in the EFTS network design. Many store-and-forward systems employ a fixed routing technique (with a small number of alternatives) without requiring the reinitialization of the net-work. This means that addresses on data streams
(routing indicators) wiiI be read at several places in the network and decisions made as to the path to be taken. This will allow EFTS to achieve the efficiencies associated with line load balancing but complicates path control with the problems inher-ent in a spill-forward system of routing, namely:
shuttling or "ring-around-the-rosey." This problem is not severe in the single regional system described by the Federal Home Loan Bank Board, but is a potential source of trouble if the fundamental net-work architecture is not sound.
3. A third consideration in the design of the access
network is link control. Link control is the ac-knowledgment of the receipt of data and the con-trol of errors on each hop through the network.
The two common types of link control are forward error correction (FEe) and Automatic Request for Retransmission (ARQ). In FEC, sufficient redun-dancy is added to the data stream to provide an acceptable degree of assurance that errors will be corrected at the receiving end without acknowledg-ment. The .oL~RQ approach concentrates on detectioll of errors and the retransmission of a block of data when an error is detected. Synchronous Data Link Control (SDLC), a form of ARQ, is employed in devices conforming to IBM's Systems Network Architecture. A similar line discipline is emerging as an international standard and is likely'" to influ-ence the design of future EFTS links. In an EFTS system, SDLC could be used between SNA devices, e.g., between an IBM 3600 Financial System and a thrift institution servicer's machine. Because of the variety of machines employed by the thrift institution servicers, however, switching systems will probably employ the line disciplines prescribed by the manufacturers for links between the thrift im-portant, however, are standards that determine the operation of intermediate devices, such as the trans-mission control units of the switch and the thrift institution servicers and the cluster controllers of the S & L's. These controls are necessary for net-work security and to pace data through the netnet-work to assure that the available data buffer capacities are not exceeded by bursts of data.
5. Finally, end user protocols must be developed for the EFTS. These protocols, used by the thrift institutions, are essential for customer security and include authentications and special codes agreed upon by the industry. As EFTS evolves, a parallel evolution of end user protocols can also be expected to evolve to support new services offered by the thrift industry.
Figure 2 summarizes the key controls in the network access subsystem by illustrating the format of an en route message nested within the control protocols de-scribed above. While every attempt should be made to reduce the communications overhead incurred by these protocols, the EFTS users will probably find the cost of this overhead to be low in comparison with the costs associated with loss of control of the network.
Other controls such as modem synchronization and signaling through a dial-up network are considered to be within the responsibilities of the supplier of the communications link, while still other controls, such as end-to-end acknowledgment and password protection
CONTROLS INFORMATION TRANSFER ON THIS LINK • (2)
.-/'--.---j
SPECIFIES THE ULTIMATE DESTINATION OFTHETEXT~
-SPECIFIES WHAT SHOULD BE DONE WITH THE TEXT BY INTERMEDIATE DEVICES
0
- - - -
FORMAT AND MEANING AGREED UPON BY ENDbit "0"
U S E R S - - . . ®
-
bit IOn"LINK ROUTING
~;~~CREOL I TEXT ERROR
ADDRESS INDICATOR CONTROL
B ~ ----~
4B
o
PROGRAM LOADED BY NETWORK CONTROLFigure 2-Key controls on an en route transaction
are considered to be the responsibility of the thrift institution servicers themselves. These controls, while important, are outside the scope of this discussion.
Figure 3 illustrates the hierarchy of facilities that is envisioned for the Electronic Funds Transfer System.
SORTING, ROUTING, TRANSMISSION CONTROL, ACCESS MANAGEMENT, FILE ACCESSING AND TELEPROCESSING EQUIPMENT
TERMINAL CLUSTER CONTROL EQUIPMENT DATA CONCENTRATION AND MULTIPLEX EQUIPMENT
OCR'S, KEYBOARDS, PRINTERS, SCANNERS, MOT'S PLUS SIGNALLING AND HEADER CREATION EQUIPMENT-PLUS OFF LINE FUNCTIONS
Figure 3-ADP equipment hierarchy
At the Switch and at the Thrift Institution Servicers, computer installations will be required to perform the switching and communications line handling functions.
(The S"llitching function can include circuit, message, or packet switching as well as certain hybrid combina-tion of these techniques sometimes referred to as
"virtual channel switching.")
At the S & L's themselves, the ADP equipment is visualized as cluster control equipment and terminals.
Cluster controllers control terminal equipment which is either locally or remotely attached. Cluster controllers contain a pool of equipment normally associated with
"intelligent terminals" and perform a data concentra-tion funcconcentra-tion so that multiple terminals can share a communication line.
Figure 4 shows some communications options to be used in the design of the EFTS. It is reasonable to expect that much of the access network will be provided by various communications carriers under contract to the EFTS operator.
The availability of the services shown in Figure 4 and their tariffs are constantly changing so that no useful purpose would be served by repeating them in this paper. There are, however, several trends in com-munications that will affect the implementation of the Electronic Funds Transfer System, particularly the access subsystem.
VALUE ADDED
FAST DIAL IN IROTARY)
VOICE GRADE lEASED POINT-TO-POINT
SATELLITE
SLOW DIAL IN IROTARYI
/11\
/1/ \ / I I \ / I I \ 0 0 0 0
Figure 4-Some communications options
Are Computers Ready for the Checkless Society? 151
The move to digital communications
The communications industry is dominated by voice users who have traditionally been provided with switched analog circuits of approximately 3000 hz bandwidth. Digitai users had to make do with the available services and adapt them to digital use through modems. Both switched and leased services are avail-able under a variety of tariff structures. Data speeds range from the sub-voice level (50, 75; 100, 300 bits per second) through switched voice ievei 1200, 2400, and sometimes 4800 bits per second, up to 9600 bits per second that can be provided by present day modems operating over a leased, specially-conditioned voice-grade line. Higher speeds are available by leasing lines that have been derived from groups of voice channels ( 6, 12, 60 and 240 voice channeis) . New construction in the telephone system is employing digi-tal communications techniques to an ever increasing extent. This is being done because the advantages of digital voice communications (circuitry is cheaper and easier to adjust) outweigh the disadvantages (up to 64 kilobits per second are used to provide high quality voice). The move to digital voice has an important effect on the design of the EFTS network. In the next few years both switched and leased digital service, at voice channel prices, can be expected to be available between major cities in the United States and will have channel capacities of 56 kilobits per second (derived from a 64 Kb/sec voice channel).
High speed access (Tl) lines will also become avail-able for data use at speeds up to 1.544 megabits per second and groups of these lines will probably be tariffed eventually to provide service up to the 96 mega-bit rate envisioned for digital television.
The use of satellites
Leased satellite capacity is an obvious candidate for the long haul trunking portion of the National EFTS of the future but a satellite system has a unique fea-ture that may be advantageous even at the regional level. This feature is derived from an earth station's ability to "hear" everything repeated by the satellite transponder and thereby establish a fully intercon-nected or nodeless arrangement of earth stations.
This permits the design of a system that can set up and tear down point to point circuits on demand and can allow the EFTS to adapt to changing loads. Satel-lite capacity is currently available and several com-panies have planned satellite services that appear to be considerably more cost effective than equivalent ground service.
Distance sensitivity
For the past several years, the cost of providing communications service has become less sensitive to
distance; that is, the cost of long haul communications has been falling and the cost of local terminations has been rising. These costs have not yet been fully re-flected in the tariffs. The Government's encouraging of competition in public communications must inevita-bly result in tariffs that more closely reflect the costs.
This will mean that the cost (to the EFTS operator) of a transaction will be relatively insensitive to dis-tance and will be most influenced by considerations associated vTV'ith the access subsystem (terminating arrangements, terminals, subscriber loops, etc.).
Dialup and switched networks
Until now, the use of switched communications for
Until now, the use of switched communications for