Davis B. McCarn

The National Library of Medicine

Bethesda, Maryland

Evil Communications corrupt good manners
The First Epistle of Paul to the Corinthians XIII:33

The National Library of Medicine now provides the most-used, on-line bibliographic retrieval service in the world, MEDLINE (Medical Literature Analysis and Retrieval System On-Line). The service is provided through terminals in 227 institutions in the United States, Canada, England and France. Over 100,000,000 characters per month are transmitted to and from the two computers providing MEDLINE service. For a description of the overall system see the referenced article.

Since the initiation of on-line services, the NLM has been concerned with improved communications services for its users. In spite of this concern and the related management efforts involving engineering, analysis, programming, cajolery, and threats, communications remains a major problem. It is estimated that about three quarters of user difficulties derive from communications services. This is not surprising when one considers the elements of the path of the usual user:

  1. His terminal;
  2. His local telephone;
  3. His local telephone service;
  4. Local network node (mini-computer);
  5. Network supervisor (to establish a network path);
  6. Two to five network path mini-computers;
  7. A mini-computer connecting the host to the network;
  8. A communications controller connecting all lines to the central computer (host); and
  9. The central computer.

This enumeration of path elements is further complicated by the fact that between each element 1 to 2, 2 to 3,…6 to 7, all signals are converted from digital to analog and back. It sometimes seems a wonder it works at all! It certainly conforms to Rube Goldberg's dictum: "Do it the hard way." If all elements operate effectively 98% of the time, there still are enough elements so that a user would have only about an 82% chance of obtaining service. Fortunately, over most of the system, probabilities are probably higher than this. But note the word "probably", we really don't know, and this constitutes one of the major problems of running a communications based service. Clearly the availability of a service is vitally important to its managers (and users), but at present, it is practically immeasurable. One does hear about problems, but seldom about successes. Users cannot be asked to keep detailed records of both effective use of the service and failures. It seems nearly impossible to get AT&T or its subsidiaries to monitor for busy signals. Providers of network services themselves may not know when and where difficulties exist. In short, this high technology is managed and monitored by the seat of one's pants because adequate information is not available. Even when problems are reported, exceptional diagnostic skill may be required to track down the specific problem and correct it.

NLM now has connections to AT&T's Direct Distance Dial Service; the Federal Telecommunications System; the TWX network; the TYMNET of TYMSHARE, Inc.; and the ARPA network. The central processor is an IBM 370/158 running under OS/VS and the Timesharing Option for communications control. A few examples of the problems which have occurred within the last year will illustrate the range of problems which have confronted one communications user.

1. Some time ago, we were plagued with regular failures of the TYMNET; with annoying frequency, a link in the network would drop out disconnecting several users. After months of work by TYMSHARE, it was determined that transmission of 40 asterisks (*) caused Bell's 200 series modems to drop synchronization giving the appearance of a line failure. {Note: even parity ASCII is hex AA which gives a repeated dibit of either 10 or 01 depending on phase relation between modem and character to serial converter. Codex V.29 modems were vulnerable to a pattern which matched 14 successive bits of the PN-127 scrambler. /RDM}

2. Within the past two months, international carriers (IT&T, M. U. I, and RCA Globecom) decided to alter a convention and send a carriage return-line feed prior to the date-time group. A carriage return sent to the IBM computer as the first character results in an automatic disconnect. Thus, users of our service in Brazil and Australia were disconnected before they could get in.

3. Our ARPA connection is a jury-rigged set-up where NLM's computer looks like a terminal (or five terminals) to ARPANET. If the ARPANET is receiving too much material from a terminal, it sends back a control-G (Bell), but there appears to be no way to make IBM software recognize this and hold transmission until the net is ready again.

4. NLM has procured INWATS service to allow toll-free access for those users not in node cities. These INWATS lines did not provide acceptable service until NLM developed procedures for measuring the signal-to-noise ratio and presented this information to the phone company in substantiation of its claims.

5. A user in a hospital in San Mateo was unable to make ungarbled connection to the service. The terminal provider and the local telephone company both insisted their equipment was operating. A visit by a technical representative established that the coupler in the terminal was faulty.

6. Users of direct-dial complain that they get no response from the system. This is usually the result of a malfunctioning data set which can be "busied out" and repaired at leisure.

These are but a few examples of the daily problems of running an on-line service on its communications side. The IBM side is no less difficult. We have had nearly unbelievable problems on the computer side. Our first telecommunications access method (QTAM) would not communicate with IBM 2741 terminals. Our present access method (TSO-TCAM) doesn't talk to direct wired ASCII terminals. We had to get engineering help from McGill University in Canada to modify our IBM 2703 to accept 300 baud lines, and the IBM 3705 communications controller was announced without 300 baud capability. It sometimes seems as though IBM and the communications providers aren't communicating.

To quote Burns,

"But, Och! I backward cast my e' e
On prospects drear!

An' forward, tho' I canna see,
I guess an' fear!

The recent (November 15, 1973) decision by the FCC to grant a construction license to PCI and the probable entry of G. E. into the communications service business under the Tariff 260 provisions clearly means that available communications services will proliferate. GRAPHNET has been approved, and approval of Telenet is anticipated shortly, probably before publication of this paper. By mid-1975, there will be at least four nationwide networks providing commercial data communications services. Similarly, if the recent past is any indication, they will not be interconnected and each will have its own technology and protocols for connection of a host to the network. Packet switched networks are also being built in other countries; the COST II project is trying to pull together the European efforts. Some U. S. networks will connect to networks or nodes in other nations and the Post, Telephone, and Telegraph (PTT) companies are now wrestling with the problems of international data networks and on-line services. These rapid developments promise to greatly reduce the cost of data communications. The cost of computers and particularly on-line storage continues to drop at a rate of 35% per year. It is a time of turmoil and promise that defies long-term planning but requires action every day to maintain reliable service, and plan rationale migration paths in the face of explosive technological change.


DB McCarn and J. Leiter, On-line Services in Medicine and Beyond, Science 181: pp 318-324, 27 Jul 1973.