Networks, Email, and fax.
General purpose computer networks began with the ARPANET, the building of which was begun in 1969. The ARPANET was first described at a major conference in 1970, when L.G. Roberts and B.D. Wessler presented a paper entitled "Resource Sharing Computer Networks" to the Spring Joint Computer Conference. At that time, software systems were not nearly as portable as they are now, and there were major difficulties in transferring a system from one computer to another. The proposed solution was to leave the system in place and to make it possible to use it remotely from other sites. This was one of the main motivations for establishing the ARPANET. Consequently, the ARPANET emphasized remote 109-in facilities, together with facilities for transferring files from one computer to another.
The ARPANET was a resource-sharing computer network. At the time of the conference in 1970 there were 20 stations connected to it. Another network, established at about the same time, but initially for the more limited role of providing access to a proprietary time-sharing service, was the Tymnet. it was described at the Spring joint Computer Conference in 1971. The Tymnet had many innovative features, but did not receive the recognition it deserved.
The ARPANET, when first proposed, was staggering in its boldness. For example, it used leased lines of 50 kilobits/sec bandwidth. Using this bandwidth was unheard of at that time. it was reported that the telephone company at first said they had no tariff category for this, and ARPA found it necessary to apply the superior persuasive powers of the U.S. Government. No doubt, large scale resource sharing networks were bound to come in time. What Roberts and his associates did was telescope the time interval-they made the clock go round faster.
The ARPANET consisted of a network of computers known as IMPS (Interface Message Processors), used solely for switching. These computers were under central control, and their programs were loaded using the network itself. The working computers, known as hosts, were connected to the IMPS. There was thus an independent communication network with hosts attached to it. The bandwidths of the leased lines and the switching capacity of the IMPs were together fully sufficient to ensure that, under all normal traffic conditions, the transmission time across the network for a 1 kbyte packet would not exceed 0.5 second, its planned maximum value. Since the communication network was independent of the hosts, there was nothing a local site manager could do to disturb this flow.
It came as a surprise to me-no doubt I had not sufficiently taken economic considerations into account-to find that the later computer networks that grew up did not follow the same principle. Instead, they were formed by the direct interconnection of the host computers. Message switching became a part-time function of the operating system in each of the hosts; this included the forwarding of messages in transit to other hosts. The efficient working of the network was thus dependent on the state of loading of the hosts and on the good will of the site managers. If a host were taken out of service for local reasons, the network was impaired by the loss of that node, which could well be an important one. It is surprising that these networks worked as well as they did. Their performance improved as host computers became more powerful and more reliable, and as the user community began to attach increased importance to network services.
A typical site is now likely to comprise a number of computers connected by a local Ethernet. The Ethernet is connected by a bridge to the wider network, and this bridge functions much in the manner of an IMP on the ARPANET. Thus, to some extent, the situation may be said to have righted itself.
It often happens that imaginative initiatives, especially when they are on a large scale, have unforeseen outcomes. Indeed, this is one of the main reasons for conducting research in computer systems. The way in which email on the ARPANET was embraced by the user community and rapidly became of major importance, is a case in point. From the beginning, time-sharing computers had a feature whereby users might send messages to one another and the designers of the ARPANET must have realized it would be possible to send messages across the net.
However, there is no reference to this possibility in the Roberts/Wessler paper, nor is there any in a subsequent group of papers presented to the Spring joint Computer Conference in 1972. The authors may have thought that the sending of messages would have appeared to the sponsors as such a trivial operation that even mentioning it as part of the case for a major investment would be counter productive. Nevertheless, after just a few years, almost everyone connected to the ARPANET made use of the mail facilities and was enthusiastic about them. There were soon many users who rarely used any of the other facilities of the ARPANET. People in the ARPA community whose work was mainly administrative, and who traveled around the country, found email invaluable in enabling them to maintain contact with their own offices and continue with their work, wherever they happened to be. I heard a lecture to that effect at a conference in Japan in 1975. It was obvious that email on the ARPANET had changed the speaker's life. It was then that I began to appreciate that to use the ARPANET for email and nothing else was not an abuse of its facilities, but rather the contrary.
Email on the ARPANET was user friendly and it remained so for a long time. In the early 1980s, when the ARPANET had been operating for 10 years and contained about 100 nodes, I was working at the Digital Equipment Corporation in Maynard, Massachusetts, I was known on the net as wilkes@)digital. What could be simpler?
The ARPANET expanded, other networks were formed, and bridges were established between them. Email was no longer confined to the small and privileged community of ARPANET users. Inter-networking brought a host of problems. In retrospect, some of these problems were inevitable under the circumstances; others were largely man-made.
The new networks were formed of computers linked by leased lines and dial-up lines. The computers lacked the processing power necessary to handle the load imposed by the users and concurrently to give the same switching performance as the ARPANET IMPS. The leased lines were mostly of voice grade and frequently formed a bottleneck. Inevitably, delays occurred and queues built up, especially at the bridges. These problems continue to afflict email to this day.
Sometimes a message fails to be delivered for a good reason, perhaps the recipient is unknown at the destination. Most networks provide for the sending of an explanatory message to the originator under these circumstances. Unfortunately, there are real difficulties in ensuring that these messages are passed correctly across bridges from one network to another. Currently, all too frequently, messages are launched into the unknown with a significant chance that nothing more will be heard of them.
Finding the correct address to which to send a message is not straightforward. The various nets grew up with their own distinctive addressing conventions and it became necessary to impose some sort of order on them. Instead of the simple flat addressing system of the ARPANET, the Internet, as the combined system is called, has domain names and other qualifiers. The proliferation of hosts has led to their being given arbitrary and nonobvious names that are hard to discover. The domain names can be equally confusing. Sometimes the local email gurus can help, but often it is necessary to call the recipient's office and ask for his correct net address.
Finding the node name is difficult enough. Unfortunately, many email users have made matters even worse for themselves by choosing to be known by initials and nicknames. Some have been forced by their site managers to adopt opaque pseudonyms, or to abbreviate their surnames, in order to accommodate those wretched computer systems that will not handle names composed of more than eight letters. The penalty for not knowing exactly how the recipient should be addressed, or for making a minor slip, is that the message is not delivered. The solution here lies in the site managers' hands. It should always be possible to reach a user using his full name; a simple expert system should be provided to sort out all the obvious variants of first names, initials, and recorded nicknames. If this fails to identify the user unambiguously, the header of the message should be referred to the mailroom clerk. That this has not been generally done suggests that enthusiasm for expert systems among the computer community is not as widespread as it is sometimes alleged to be.
In spite of the problems spelled out here, the present system of interconnected networks does succeed in handling a great deal of traffic. It must be remembered that this system has been put together by universities and research organizations on limited budgets and it is available to the ordinary user free of charge. Efforts are being made to improve the quality of the service and improvements are occurring steadily. In the meantime, a powerful competitor in the form of fax has appeared.
Fax is based on the telephone system. The steady improvement of that system-internationally it has been a dramatic improvement-has made fax possible. Equally important have been developments in modems. The most advanced are capable of sending at a speed of 9600 baud if the line is good enough and of automatically choosing a lower speed if it is not. For a time, fax was handicapped by a lack of accepted transmission standards, but this is no longer the case.
Fax is compatible with normal office routines, and is easily operated by office staff. With email it is first necessary to teach people how to use a computer. A fax message is addressed to the recipient by name and sent to a fax machine at his or her office; the office staff sees that it gets to the recipient. Finding the fax number is just like finding any other telephone number.
If the distant fax machine is busy, there will be some delay in setting up the connection. Once this has been achieved, however, the message goes through in real-time and one can be sure that it has arrived. There is no storage in the network; any necessary queueing takes place at the sending office.
The telephone network is suited to the transmission of messages because it is universal, it has a well developed addressing system that works, its bandwidth is adequate, and, from the point of view of the user, messages go directly to their destinations. The telephone company, rather than the user has all the problems associated with routing, redundancy, and upgrading of plant.
Computer networks serving groups of users with a common interest are suitable for remote login and for the transfer of files that are too long to be sent as messages. Such networks provide efficient communication between working partners, they can have a very wide bandwidth, and they enable economic use to be made of leased lines. No one doubts that such networks are needed or that they are of great importance. McCarthy questions, in my view very pertinently, whether the computer community should depend on them for email or whether email should, instead, be sent via the telephone system as fax is. He makes various suggestions about how the telephone system for email use should be organized.
Fax depends on the transmission of scanned images and email depends on the transmission of ASCII characters. These appear to be quite distinct techniques. Nevertheless, we may expect to see them converge. Already, some fax machines will accept a message in ASCII form from a local computer network and transmit it as a scanned image. It is a small step to design fax machines that will transmit messages in ASCII form when the machine at the other end can accept them. McCarthy mentions this possibility. On the computer side, the transmission and storage of scanned images is a topic on which much research is being done. At the more practical level, there are already available fax server cards that can be inserted into a personal computer or a workstation.
If and when the convergence has taken place, a user working at a terminal or at a workstation will have the option of sending a mail message either via a dialed connection or via the network. To be safe, he would, under existing conditions, probably choose the former, unless there were no bridges to be passed and his level of trust in the network were high. On the other hand, if his message happened to be of no great urgency and he wished to avoid the charges associated with a dialed connection, he would perhaps opt for the network. The decision as to whether the message should be sent as a scanned image or in ASCII form is one that he would normally leave to the system.
I have stressed that with fax there is no queueing in the network. A message remains at the sending end until a connection to the recipient can be established and transmission then takes place in real-time. Some computer networks, for example DECNET, have the same property. The fact that there is no queueing in the network does not prevent delay from occurring when there is congestion at the destination, but it does ensure that the delay is visible at once to the sender, who can consider what alternative courses are open to him.
It can be argued that a delay of an hour or two is of no great significance. The difficulty is that once delays are admitted, there is no known effective way of putting an upper limit on them. In my opinion, no email system in which messages are subject to indefinite and perhaps lengthy delay can hope to compete with fax in the engineering and business worlds.
Maurice V. Wilkes received the ACM Turing Award in 1967 and is the author of Memoirs of a Computer Pioneer, MIT Press, 1985.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Computing Perspectives|
|Author:||Wilkes, Maurice V.|
|Publication:||Communications of the ACM|
|Date:||Jun 1, 1990|
|Previous Article:||ACM Forum.|
|Next Article:||Evaluation of safety-critical software.|