VII. Network Architecture
In the present section we will summarize the main points that we have made so far. The subject is moderately complex, and the relevant technology is evolving very rapidly. This makes it difficult for most customers to keep up with new products and even taxes the abilities of suppliers' representatives to do so. The present paper hopes to provide a school district's technical staff with the vocabulary necessary to negotiate the purchase of equipment to connect their schools to the Internet and to provide for the continued maintenance of this networking equipment.
The success of the Internet has been due in no small measure to the fact that it uses a standard public protocol. This enforces interoperability among the millions of devices connected to the Internet. At another level the Internet has enforced a uniform networking architecture, which we endorse and hope to clarify in the course of this paper. We believe that this architecture is likely to provide a solid and enduring foundation for national and international networks which serve education, government, business and entertainment.
The key elements of the architecture that we have been discussing are the local area network, the metropolitan area network and the wide area network. In the subsections which follow we elaborate on these points. Our discussion of the local area network emphasizes the interoperability of devices on the network. As for the metropolitan area network, we are more concerned with how school sites can connect to this network than with its overall structure and content. And with regard to the wide area network, we reiterate simply the virtues of packet-based services.
The summary in the paragraphs which follow provides a useful basis for the more speculative discussions of section VIII, which deal with services either still under development or in the earliest stages of deployment. Since any school network should be designed to last for a number of years, it is important to keep these horizons in mind as one proceeds, even if near-term purchases will involve more familiar technologies and devices.
Within each school we advocate a local area network capable of transporting IP packets. The physical medium for this network would typically involve twisted pairs of copper wire, similar to that used in standard telephone installations and hence familiar to the personnel who install and maintain current school telephone systems. Another option uses coaxial cable similar to that employed for cable television hookups. A common implementation which provides speeds of 10 million bps is known as Ethernet. Large sites may require a fiber optic backbone to tie together a number of individual Ethernets, the total reach of any given Ethernet being somewhat limited. Typical costs for Ethernet wiring are in the range of $100-$350 for each potential network attachment point. These costs are hard to state in general, since they are strongly affected by local considerations such as labor rules, asbestos abatement requirements, or unusual structural configurations. There is an added cost for the electronics associated with each attached device; this adds another $150 to the cost of each device.
Devices on the LAN can include a large variety of computers. Most personal computers currently on the market allow for attachment to an Ethernet, often with an add-on card priced at about $100 and increasingly with a built-in Ethernet interface. The same is true for all high-end workstations and for many printers and other peripherals. The school's mail server, file server and information server could either be a workstation of this type or a high-end personal computer equipped with enough disk space and internal memory to perform the required tasks.
The other key device on the network is a router, which examines packets flowing on the local network and selects those with external addresses for routing through the metropolitan area network out to the wide area network. We advocate using a separate router for maximal network stability and flexibility, but the same functions can be provided by the school's server or by a specially configured personal computer.
The link to the metropolitan area network might seem to be a minor component of the network architecture for school networking. It turns out, however, that this local loop (to use the jargon of the telephone company) can be one of the most costly components of the whole network infrastructure. The reason for this is that its installation and maintenance are labor-intensive, and it is a facility, unlike other components of the network, which isn't shared with any other users. Hence it is very important to select the technology for the local loop with considerable care.
Obviously the local loop is an important component with regard to the level of connectivity enjoyed by any given site. It is this link which determines the site's bandwidth to the outside world. And the level of reliability of this link is what - more than anything else - determines the level of reliability of the overall networking scheme.
In the preceding sections we have discussed several technologies which can be employed for the local loop. These included analog telephone service, digital ISDN service and private leased lines. In the following section we will review these options and expand upon them to include several new and promising possibilities. A typical school site might employ more than one of these technologies: a high-speed link could be used as the primary means of connectivity, with fall-back to lower-speed options in case of problems with the primary link. This type of redundancy increases the reliability of the school's network connection and allows for remote diagnosis of problems when they do occur.
The connection of the school LAN to the Internet Cloud is a packet-based connection. We do not believe that individual dialup access is a viable means of providing Internet services for whole school sites. Packet-based services allow for efficient multiplexing and resource sharing that make for significant cost reductions.
One can compare the model of dialup access with that of packet-based services and find some startling differences. Consider, for example, a 2400 baud dialup to an on-line service billed at $10 an hour. This is representative of neither the cheapest nor the most expensive of such services. Contrast that with a university's 1.5 million bps connection to the Internet, which might be priced at $20,000 annually. The $20,000 annual fee works out to $2.28 an hour, which is already cheaper than the dialup service. But the bandwidth of the university's connection is 643 times that of the dialup link. The overall cost advantage of the university's connection is a factor of about 2500.
Although these figures point the general direction in which to go, namely toward packet-based connections and packet-based services, there are still many possible ways to implement such an architecture. The demands of a growing National Information Infrastructure will push the development of technologies which serve the needs outlined n the present paper. In the following section we list some of the new technologies which appear most promising and try to guess which of them will be most suitable for school networking applications over the next few years.