BYTE, October 1994
Whether fronted by a character-mode or a graphical interface, or whether built on a mainframe, a Unix server, or a stand-alone PC, traditional on-line services and WANs share a basic assumption: Bits travel from one end to the other over a dumb pipe. The minicomputers at the core of CompuServe, the desktop PCs and Macs calling America Online, and the laptops dialing into remote-access servers have all the brains, while the networks to which they attach know virtually nothing--other than the addresses of the sender and recipient--about the messages crossing their wires and waves.
All that, however, is about to change, as a new generation of "smart" networks comes on-line. Drawing on the efforts of thousands of researchers and programmers in the telco, Internet, and on-line communities, General Magic (Mountain View, CA), AT&T (Basking Ridge, NJ), and IBM (Armonk, NY) are developing network services that will change the very definition of a WAN. Instead of being a mere conduit for applications executing at its endpoints, the network becomes a host for distributed applications that execute inside and outside the "cloud," or perimeter, of the wide-area service. Sun Microsystems got it right years ago when it prophesied that "The network is the computer."
The end-user device connected to the smart WAN can be simple or state-of-the-art, stationary or mobile. The network can adjust itself to the access device, and an application assumes the presence of a flexible communications infrastructure. The user and applications developer are shielded from the ugly complexity of networking and communications, and service providers can reach new audiences without tailoring their offerings to a specific delivery platform.
Instead of relying only on real-time, connection-oriented sessions, smart networks make extensive use of store-and-forward messaging transports. They are designed to host software agents, or proxies, that move around the network, routing or filtering messages sent to a user and seeking out information or services on the user's behalf. Agents can work all the time, even when users are asleep.
The advantages are powerful: Agents make data networks smarter about people, instead of requiring people to be smart about networks. They let you focus on getting your job done, rather than on the details of how to communicate. You can get the messages you want wherever you are, prioritize your work, and eliminate the chaff. And agents help you find information or services without having to know about and delve into every corner of the network.
The first commercially available agent-based WAN, from AT&T, is slated to come on-line this fall, but it will take years before smart networks are common. In the near term, their growth will be limited by their novelty--especially by misunderstanding and mistrust of agents--and by the cost of switching from existing solutions. However, over time, agents will be an essential component of networks because they provide a flexible means of accommodating the exploding variety of devices and services.
The emergence of the intelligent backbone is occuring in parallel with advancing capabilities at the endpoints of the network, such as LAN server-based telephony services, desktop telephony APIs, and more communications-enabled applications (see "Computer Telephony," July BYTE). In fact, a rivalry is shaping up among players (e.g., computer and peripheral makers, software companies, and some phone equipment suppliers) who build advanced telephony products for the desktop and those companies (e.g., carriers, packet-network services, and some equipment suppliers) who want to push intelligence into the network itself and profit from increased fees and usage. These approaches are not mutually exclusive; some companies, such as AT&T, straddle both camps and stand to benefit in either scenario.
For example, one emerging software category is the "universal inbox," a single place where faxes, E-mail, and voice messages are collected and presented to the user. Apple first implemented such a capability in System 7 Pro, and Microsoft is building one into Chicago that, if used with services that support Windows TAPI (Telephony API), will handle all these message types. Third-party products are also available, such as OneView from Centigram Communications (San Jose, CA). But doing this task on the desktop is only one approach: Carriers aim to support it as well in their networks, making messages accessible to a user from work, home, or on the road.
No matter how widespread client-side telephony applications become, smart backbones are bound to flourish because of inherent technology and market shifts. One driving factor is the transition to mobility and wireless networking. Wireless links are more expensive, less reliable, and slower than land lines, so real-time connections are harder to maintain, and the exchange of rich media is impractical. Good wireless protocols tolerate interruptions and quality degradation and are designed to connect, exchange brief messages, and sign off; store-and-forward messaging is thus the best solution for intermittently connected computing. Mobile devices, with their premium on size, weight, battery life, and low cost, also aren't well-suited to resource- and bandwidth-hungry RPC (remote procedure call) mechanisms. When you add the complexity (from a service provider's point of view) of supporting multiple devices (e.g., a laptop, an organizer, a pager, and a cellular phone) per user, there's an obvious need for lighter-weight and more flexible architectures than traditional terminal processes, RPCs, and virtual circuits.
This is where agent-based networks shine, because they work well on connectionless, low-bandwidth infrastructures. Agents facilitate mobile attachment, relieve work from remote systems (permitting them to be smaller and cheaper), and simplify interfacing with multiple devices and back-end services. Traditional WANs let you accomplish work only when you're connected or, if responsibility is pushed out to the client, require data reconciliation during connect sessions. Agent-based WANs, by contrast, let you inject a task into the network that executes whether or not you are connected. The agent operates continually, as near to the data as possible, minimizing network traffic and reducing how much the user's system and the back-end service need to know about each other. When it has results to report or needs further guidance, the agent finds its way back to the sender.
A classic example is E-mail filtering and routing. In a client-based implementation, the rules engine lives on your system, not on the network, so messages aren't processed or screened unless you are attached to your mailbox. In an agent implementation, the rules execute on the network when you're not connected, and your mail filter can even contact you via pager if you receive an urgent message. This scenario could be implemented as a conventional server process, but it would be network-specific and have to be programmed in advance. Agents permit greater flexibility; for instance, they could teach a server new message-handling techniques by injecting a mail-filtering method.
AT&T's PersonaLink and the as-yet unnamed offering from IBM Intelligent Communications Services are new agent-based systems. These services use different architectures from one another and aim at widely different audiences and applications; however, both provide an agent-based message-handling and information-delivery platform.
AT&T's PersonaLink, built around General Magic's Telescript technology, is an electronic community for E-mail, information retrieval, and on-line shopping slated to become available this fall in the U.S. only (for more on Telescript, see "Agents Away," May BYTE). The network consists initially of centralized servers accessed by 800-number service or wirelessly through the Ardis packet radio network underneath Motorola's MNI (Mobile Network Integration) service umbrella. Eventually AT&T will add local-access nodes on a nationwide packet network, expand to distributed servers, and add service in other countries.
Using PersonaLink requires Telescript-enabled devices and "Telescripted" software. At first, this means that only two devices will work with PersonaLink: the Motorola Envoy and the Sony Magic Link. Both are handheld computers (or Personal Intelligent Communicators) designed around General Magic's Magic Cap operating system, which has Telescript built in. Additional Magic Cap devices are expected to be introduced in 1995 by General Magic alliance members Philips and Matsushita, and perhaps eventually by Apple or AT&T. Also in early 1995, General Magic is expected to release Magic Cap software for Windows and the Mac OS, which will make it possible for Macs and DOS/Windows machines to talk to PersonaLink.
IBM's Intelligent Communications service, slated to start up in 1995, is a different beast. It will be a communications "super-service," a hub for routing and translating communications from one service and medium to another: desktop to mobile, PDA (personal digital assistant) to mainframe, E-mail to fax, and text to speech. For instance, once it is fully implemented, you might send a cc:Mail message into the service from your desktop, use a wireless Newton in the back of a cab to view a fax, and then have the network read your IBM PROFS mail to you over a cellular phone. The sophisticated routing features would let you specify a scenario, such as "If I get a fax from Toshi regarding the Kyoto project, please run it through OCR and read it into my voice-mail box, send a copy of the text to Barbara, and forward the fax image to the optical-archive mailbox in the legal department."
Outside the Intelligent Communications cloud, IBM doesn't require new devices or protocols or applications; on the contrary, it welcomes and supports virtually any communications technology now in use. But inside the cloud, it uses an innovative agent architecture that isolates subscriber profiles and preferences from access devices and service providers, all under the aegis of centralized billing, security, and authentication.
These two services symbolize different concepts of the agent network. PersonaLink is an applications environment, a built-from-scratch messaging platform through which third-party content and service providers can deliver information, entertainment, and shopping. It is aimed at individual consumers, especially because it will be reached initially through the user-friendly Magic Cap interface. It's not meant as a message gateway, although it will support Internet and X.400 mail, fax, and paging. And while Telescript could someday become the lingua franca of wide-area communications, PersonaLink does not now support a diversity of access devices.
By contrast, Intelligent Communications is an umbrella for smart message routing, a giant gateway for consolidating wide-area communications among mobile professionals. It will relieve corporations of the considerable burden of creating and maintaining their own networks, but it's not an applications environment per se. User and service programs still execute outside the cloud, but they can become network-enabled by supporting Intelligent Communications APIs and object standards.
In a sense, PersonaLink takes the longer view, to a future where software agents act out human wishes; it provides a marketplace for agents or a framework for negotiation between customers and suppliers. Intelligent Communications is more about the here and now, addressing the frustrations and desires of today's mobile and on-line users. However, its state-of-the-art solution is designed with a clear eye to the future.
When IBM set out to design Intelligent Communications, it surveyed mobile users about their problems and wishes and assembled from these a set of objectives for the service. The studies showed that users had trouble with "roaming" among carriers and desired a seamless network with a single point of customer support and one bill. They also disliked having to check multiple inboxes for voice mail, public and private E-mail, and faxes. They complained of information overload and asked for new ways to filter and prioritize messages and for a means of event notification for important messages.
The key attributes of IBM's solution, says Doug Sweeny, the general manager for Intelligent Communications, are personalization and integration. The service lets you tie together everything you're already using, create custom message-processing scenarios, and wrap it all in the interface of your choice. You can also receive information (e.g., headlines, weather, and stock quotes) from third parties in the form of messages sent to your inbox. "We use intelligence to mask the complexity of the network, and we use agents to help you gather, prioritize, and automate handling of messages," Sweeny says.
In AT&T's PersonaLink, all devices and programs must be tooled to speak Telescript; in Intelligent Communications, they speak in their native vocabularies, and the cloud translates for them. And instead of opting for the lowest common denominator, these translations are aimed at the highest capability level supported on each platform. Thus, a rich text document might be converted to simple text for display on a pager's LCD but could appear as-is on a graphical Newton.
Intelligent Communications uses a modern, message-based, interprocess communication architecture in which tasks are insulated from one another by published interfaces. As in contemporary system object frameworks, no direct "hard-wired" interaction occurs between processes; rather, their communications with one another are passed through an arbitrating layer or channel. This software layering lets Intelligent Communications achieve its goal of total neutrality.
The service is neutral with respect to devices: Any manufacturer's products can work with it, and all services are available to all devices within their inherent limitations (e.g., a cellular phone cannot display faxes). By separating presentation from service, the IBM network permits users to continue using their current devices, environments, and applications. It is also neutral with respect to carriers, media (land line, cellular, CDPD, X.25, or ISDN; low-bandwidth or intermittent), and protocols. Carriers can maintain their own directory services or systems management capabilities without affecting the behavior of IBM's service.
At the back end, the same kind of abstraction applies. Service and content providers can join the network with their existing offerings, which are virtualized to appear as Intelligent Communications objects, or they can retool their service interfaces to accept and emit Intelligent Communications objects. As long as the service provider is willing to bind to an enabling library or to support emerging standards such as OLE or OpenDoc, users can tap into their services without starting from scratch.
Intelligent Communications will run over a variety of media, including Ardis packet radio, the joint IBM/Sears Advantis network, and (unannounced) third-party phone and packet networks. It will offer gateways to Prodigy, the Internet, and most other public and private E-mail systems. The core hardware servers are IBM RS/6000s running AIX (just like the Internet backbone that IBM co-operates under contract to the federal government), and the basic internal network protocol is TCP/IP. Intelligent Communications uses X.500 directory services internally but will, in principle, be able to make use of external directories (i.e., in enterprise or carrier systems) as well as to publish its own directories to these external systems. The service will accommodate two kinds of network traffic in its first release: analog voice and asynchronous data.
The key to making Intelligent Communications work is its use of proxies, or agents, that stand in for subscribers, devices, mailboxes, and services (see the figure "Intelligent Communications"). The network never connects subscribers directly to services, because, IBM says, this can force application-specific coupling between the front end and back end, inhibit support for transparent mobility, and limit opportunities for adding intelligence (and value) to a subscriber's use of the service. Likewise, directory inquiries never discover the actual network address of a service; instead, all such interaction is arbitrated through the cloud. As a result, users never have to know how or where to find a service, and service providers never have to understand how to interact directly with a subscriber unit.
Inside the cloud, "the currency of the network is objects," says Mac McInerney, assistant general manager of development for the service. Intelligent Communications objects are self-descriptive, containing both static data and dynamic scripts; their presentation and interface elements are highly abstracted so that the information they contain can be adapted to the context in which it is being presented. Therefore, a list box on a laptop GUI could convert automatically to a spoken selection menu in a telephony interface.
The subscriber proxy virtualizes the characteristics and preferences of a user. Every subscriber (including services) has an entry in the distributed X.500 directory, including name, phone and fax numbers, physical address, and home location (home location is the default service node where a user attaches to the network). Mobile users can also shadow their home location in another place for faster response times. Because the network is distributed, your subscriber proxy is available to represent you anywhere you log in.
Information about the device or devices you use to attach to the network is contained in device proxies, which are independent of the subscriber proxy. Thus, the same person can connect at one time with a PC and at another time with a telephone, and Intelligent Communications will accommodate the difference. Even endpoints that have been tooled to directly accept and emit objects talk through "lightweight" proxies that administer functions such as security and billing.
Device proxies hold information about each machine's ability to present various objects, knowing, for instance, that one system has a color screen and keyboard while another has a monochrome screen and stylus. Devices can limit the abilities they expose at any given time, so if you wanted to receive your E-mail in spoken form while driving, your PDA could represent itself for the time being as speech-only. These proxies also understand network access methods, so delivery can be optimized to the medium being used; thus, you could automatically postpone transmission of a video clip during a wireless session until you were later connected over a T1 link.
It is in conjunction with the device proxy that IBM's remarkable format translations occur. For now, these include text-to-fax, fax-to-text (via OCR), and E-mail and document format conversions. Down the road, IBM plans to support a speech-based user interface and will add speech recognition, as well as text-to-speech and speech-to-text transformations.
The Alter Ego is a programmed rules engine that represents a user's routing preferences (including how accessible the user wants to be). Over time, IBM says, this will evolve into an adaptive inference engine that learns through observation how you work and what you like. The first level of the engine does fast routing of real-time interactions, such as phone calls and paging. The second, deeper level handles nonreal-time events in negotiation with other network services or Alter Egos; for instance, following the instructions "If I get mail from the CEO, page me; if I get any faxes from Finland, forward them to my hotel in Orlando; send all other mail to Murray," Murray's Alter Ego might say, "If I get mail forwarded to me from anybody, put it into my low-priority folder."
The service proxy is, in some senses, the most important of all, because it is through this mechanism that IBM hopes to accommodate its huge mainframe customer base and the great majority of information databases that now use legacy systems and software. The service proxy virtualizes the back end, making it appear to the network like an Intelligent Communications object. This way, you can mask a programmatic API (e.g., use the proxy to issue a SQL query against a DB2 database) or even a terminal interface. The service proxy can also serve as a gateway to LANs.
Service proxies know about service addresses, network types, protocols, and billing systems, yet they hide these details from a user. "Intelligent Communications-adapted" services retain their native interfaces and rely on the proxy for translation to and from the object framework. "Intelligent Communications-enabled" services go a step further, tailoring their front ends to accept and emit Intelligent Communications objects. One advantage for the vendor in full-enabling is that it may reduce network traffic between the service platform and the network.
Intelligent Communications supports legacy services with a proxy that acts as a terminal emulator, operating a session between only itself and the host. By encapsulating the entire legacy system in an object wrapper, IBM allows traditional information providers to offer services on the network without immediately retooling their interfaces or downsizing to servers. The service proxy also solves potential timing problems, because real-time communication occurs between only the proxy and the service, not end to end from the subscriber device to the host. This decoupling frees a user to access session-oriented services over less predictable wireless transports.
The real advantage of the service proxy is that it lets providers, even small outfits lacking programming expertise and resources, get onto the network easily. AT&T wants to encourage similar participation with PersonaLink, but making the leap to Telescript requires more up-front investment. After all, how many neighborhood flower shops or pizza parlors will be able to set up and maintain an on-line, object-oriented merchandising server?
From the very beginning, AT&T has viewed PersonaLink as both a toehold into future business opportunities and a test bed for emerging network concepts. Says Gordon Bridge, president of AT&T EasyLink Services, the division that includes PersonaLink: "This was our first opportunity to develop an end-to-end solution based on connectivity--not piecemeal, not retrofit, but from scratch." What this includes, Bridge notes, is a new operating system, new devices, a new network, new customer-service programs, and new billing services--in short, "every aspect of the infrastructure is green field."
AT&T took this risk in part because it wanted to explore and validate new ideas in network management and business organization, something too dangerous and expensive to undertake on the existing phone and data networks. "It's very hard to change a service once it's up and running, so we architected PersonaLink the way it ought to be done from the beginning," says Joe Gigas, network operations manager.
For reliability, the service uses features such as redundancy, self-healing, and performance-based monitoring (i.e., setting targets based on actual user throughput and capacity, instead of arbitrary internal factors such as disk utilization or error rates). And, in a break with the past, PersonaLink uses almost entirely standards-based hardware and software management (e.g., SNMP and HP OpenView). As a result, it requires much less human supervision than traditional telco networks and can run on less expensive off-the-shelf equipment. Much of the internal operations management is performed via Telescript agents that query services and report results. "This is a model for future network design and management," Gigas says.
In its initial release, PersonaLink is centralized, but it's designed to be scalable and distributed. Built on RISC-based symmetric multiprocessor servers running Unix, the service operates out of four data centers (one for the core Telescript engines, and one each for network operations, customer support, and billing) connected by T1 (private, 1.544-Mbps) data lines. Incoming calls to the 800-service or from Motorola's MNI are routed to a terminal server/router and from there across Ethernet to the server core. In the initial release, all third-party services, such as shopping and information, are centralized at AT&T, but in the future, third-party Telescript engines can run anywhere.
Aside from its genetic use of agents, PersonaLink is different from previous services in several ways. It's designed to support multimedia message types, including voice, graphics, and annotation, and, of course, it supports intelligent routing and smart mailboxes. All addressing is done with human names, not strange numbers or codes, and it uses advanced security provisions, including automatic encryption of all communications.
A typical PersonaLink scenario begins on a hand-held device running the Magic Cap operating system. To create a message, the user selects a recipient from a local address book or queries PersonaLink for an address and then writes the message using a local text processor. Magic Cap includes a drawer of "rubber stamps" for messages (e.g., Urgent), which appear as icons on the message as well as translate into Telescript handling instructions.
The finished message is combined with a script that tells it how and where to go, thus becoming a mobile Telescript agent. The script starts executing locally because it could include instructions for, say, accessing the local file store. But when the Telescript engine encounters a Go operation, it stops executing the script and readies the agent for transmission by saving all the variables, pointers, and stack values. If the place to which the agent is going is on the local machine, it is sent there; otherwise, it's wrapped up in a message, encrypted by a local RSA-like public key encryption routine, and dispatched across the WAN.
This "bag of bits," as AT&T calls it, travels to its destination over a live connection. When it arrives at the PersonaLink service node, it passes through a gateway process where it is immediately dispatched to a dedicated authentication and security server (see the figure "AT&T PersonaLink"). PersonaLink uses two-way authentication (unusual in on-line services), in which both the device and the service must prove to each other that they are legimate. Rogue devices are thus prevented from logging on and, conversely, users can't dial in to a phantom service that pretends to be PersonaLink and steals personal data, mail, or merchandise orders. The authentication server also decrypts the message.
Once past this security barrier, the agent is copied to a file store for backup, and only at this point is the connection with the sending device broken. The device can then carry on with other activities while the agent does its job at the PersonaLink service.
First, the agent goes to a node transport subsystem, which looks up the recipient's mailbox address in the Finder database. It then goes to the mailbox (itself a stationary Telescript agent), which can decide whether or not it wants to accept the agent. If it does, the message is unwrapped, and the script continues to execute.
Even at this point, no interaction has occurred between the mobile and stationary agents. Only when the Telescript engine encounters a Meet command in the mobile agent do transactions (e.g., mail delivery) occur between the two. Of course, interactions between agents can be used for applications beyond E-mail. For example, a user on a device could send an agent to pick from the available seats at a concert or to pick a book from an on-line catalog.
In the first release of PersonaLink, the smart mailbox will allow users to filter and route incoming messages based on sender or contents, to auto-forward messages to another mailbox or fax machine, to autodelete messages, or to be notified via pager that a message has arrived. Mead Data Central will provide a news feed (i.e., headlines, stocks, sports, and weather). A shopping "mall" from start-up eShop (San Mateo, CA), which will front other electronic merchants still to be announced, will appear shortly after the initial release date.
PersonaLink's software architecture allows for a mixture of data sources; for instance, to reduce data transmission needs, an electronic catalog, including images, graphics primitives, templates, and parts lists, could be shipped to users on a disk or PCMCIA card. Then the only data that would be necessary to send over the line would be new items, current prices, and spot sales. Another way Telescript preserves bandwidth is by exchanging only "reference pointers" to objects that already exist on the target platform. For instance, most icons exist in firmware on Magic Cap devices, so if they're used in an incoming message, they can be referenced instead of being sent from the server.
In adherence with a "shoppers bill of rights," AT&T won't provide back-end vendors with information about individual users. But PersonaLink will be able to monitor behavior such as frequent trips by a user to a given store. Then, when the user returns to that store, the network could flag this to the vendor, who might decide to offer an on-the-spot discount or some other purchase incentive. (The only way a vendor will be able to "acquire" a customer name is if a user buys something from a vendor, just as it is today in the real world.)
The ultimate benefit of Telescript's agent architecture to users and vendors (and to AT&T) is the flexibility it affords. Contemporary on-line services are generally monolithic: To add a new feature to AOL or Prodigy requires that it be written into the core, carefully tested, and brought on-line without disrupting 24-hour access. PersonaLink is also aiming for 24-hour service, but its design allows agents to teach the network new tricks all the time. "We didn't have to build in all the capabilities ourselves," says Alex Gillon, director of technology development for PersonaLink Services. "This has huge implications for network design and management, because we don't have to keep turning over the network software."
PersonaLink's open design could unleash a cottage industry of agents: Perhaps someone will build a better mailbox or devise an agent to find Rolling Stones tickets. But the service also needs to be protected from disruptive or merely badly written agents; hence the major emphasis on security provisions (see the text box "Telescript Security").
AT&T's PersonaLink and IBM's Intelligent Communications share their use of agents, their openness, and their enormous ambitiousness. IBM is looking for a way to keep its bread-and-butter large accounts in the fold, and Intelligent Communications will offer these customers a way to enable legacy applications with mobile access. Rather than see customers drift away to distributed computing solutions from other vendors, IBM is providing the much-needed glue to pull together public and private E-mail, text, image, voice, and paging under a single umbrella--even if some or all of these services run on non-IBM platforms. While it's refreshingly ecumenical, Intelligent Communications is deeply based on a familiar IBM concept: Some people will pay good money to have you take responsibility out of their hands and provide a bulletproof solution.
AT&T's strategy is different: The company doesn't have a large base of business-computing users to protect, but it does dominate the domestic market for voice and data telephony among companies and consumers. PersonaLink is a framework for the future of wide-area telecommunications, whether it's for the purpose of calling your grandmother, negotiating a business deal, or buying a wool sweater. It has the potential of becoming the supermall of the information highway, the on-line service for the 180-million Americans not now on-line, but it could also form the basis for business-to-business electronic commerce.
As with several projects in the telco giant's past, however, PersonaLink betrays an AT&T-centric worldview. The service is all new, requiring users and third-party vendors to start from scratch with new devices, new software, a new interface, and a new service subscription. In effect, AT&T is saying, "If everybody would just convert to our standard, we could all communicate." On the other hand, IBM is saying, "Come one, come all; we'll translate among your formats--and we'll charge you for the privilege."
AT&T argues that having once made the investment to leap into Telescript, users and vendors will discover richer on-line services and new ways of conducting business. Given that Telescript is not proprietary technology, AT&T argues that it will succeed because it is the first to market and will be the best at running a network.
Perhaps AT&T's biggest potential stumbling block is this conceptual contradiction: PersonaLink is aimed at consumers and individuals and, for now, is accessed only from the cartoon-like Magic Cap interface. Yet the access devices (e.g., the $1500 Motorola Envoy) are hardly priced at consumer levels. Resolving this conflict will have to be a priority for AT&T if it is going to attract a large enough base of users and a community of information and merchandise vendors.
IBM's dilemma will be that in trying to please everybody, it may be biting off more than it can chew. Its Intelligent Communications service is already running a little late and isn't expected to be fully up and running until late in 1995. Meanwhile, other services will continue to add gateways and format converters. If you add to these a universal mailbox and mail filter on your client, you can get much of what Intelligent Communications aims to provide but without ongoing charges.
Agents aren't just a neat technical breakthrough that simplify mobile access and enable new types of back-end services. To an increasing degree, they will become an essential aspect of WAN design because they handle ever-increasing complexity. Service providers can't possibly support the surging base of users, devices, and on-line vendors with monolithic system architectures. Adding more and more point-to-point gateways or rewriting their service's core engine to accommodate new features is economically infeasible. So the only choice is to embrace the new world of distributed, modular, object-oriented technologies.
The Generations of On-line ServicesThe evolution of on-line services has mirrored general computer technology, from host-based to client/server, and from character-based to GUI and now to agents. In first-generation services, information came from the host, and all the work was done there except for local screen-rendering and locally maintained address books, message folders, and downloaded files. Second-generation services introduced GUIs and the exchange of graphics primitives; more computer power was presumed to be on the user's side of the pipe, and the architecture shifted subtly toward client/server. In agent-based services, roving software objects can conduct business on behalf of users even when they aren't connected to the network.
IBM's Intelligent CommunicationsIBM's Intelligent Communications is a messaging hub that seamlessly connects virtually any subscriber device to any back-end service, translating among formats as necessary. Its key architectural component is proxies, or software agents, that abstract the characteristics and preferences of users, devices, routing paths, and service providers. Proxies will let you access a LAN-based E-mail system from your notebook at one moment and later have a host-based news feed read to you over a cellular phone.
AT&T's PersonaLinkAT&T's PersonaLink now employs a centralized architecture, but it's designed to be distributed. Users connect through gateways from 800-number or wireless services and are immediately authenticated. Only then are their agents accepted and passed through a node transport subsystem to an appropriate Telescript engine. The PersonaLink customer-support, billing, and operations centers are separate entities from the service core. PersonaLink will have mail gateways to AT&T Mail and from there out to X.400 and the Internet. Mead Data Central will provide a news feed to PersonaLink subscribers.