How Not to Become a Billionaire

Nobilangelo Ceramalus tells a rueful story of how the modern Seven Labours of Hercules failed to clean an Augean Stables of bureaucratic inertia and stupidity

Long ago, in the late twentieth century (1983, to be precise) I had an idea. I thought it was original, because I did not know what others were doing. The substance of it will become clear shortly, but what sparked it off was that I wanted a microcomputer, specifically an Acorn BBC B, because I had looked at what was available and decided that that was the best. But I could not afford one.

So I called the big television-rental companies in New Zealand and suggested that they consider renting out microcomputers too. In my efforts to persuade them I said that they could also sell software and such over the telephone lines, and provide information access. I did not say that I did not know of anywhere where these things were actually being done, but my computer background told me that they were possible, so it seemed a practicable idea. But the companies said that they had looked at micros and decided there was not enough of a market. My idea did not impress them at all.

A few days later I was catching a bus from the city and had some time to kill while I waited, so I wandered into the adjacent stationer. My eye was caught by a UK magazine called Which Micro?, published by a company in London run by Richard Hease.

To my astonishment it described a joint-venture called Micronet between Richard Hease's company, a modem manufacturer that he had helped set up, and British Telecom. Micronet was a first, making it possible to connect home computers over the telephone lines to a central database of information and software. The communication and display protocols were videotex (which looks the same as teletext), but its philosophy and use of personal computers as terminals made it the forerunner of today's Internet. The article also described facilities such as downloading software.

"Good grief!" I said to myself. "That's my idea."

I had not been original after all. But I could be quick off the mark. So that evening, when the clock came round to the right time, I called Richard Hease and asked if I could be his New Zealand agent. That was a Monday. He asked me to find out how many micros there were in New Zealand and call him back. I invented a method of working that out, by doing wholesale and retail surveys that gave me approximate percentages of the market for each brand, combined that with accurate unit-sales figures for one or two brands, double-checked with different sources, and arrived at a total of about 10,000.

I called Richard on the Friday. He said that that was enough for a market-base so, yes, I was now his New Zealand agent. He also told me that he had had a couple of chaps in his office the previous day from a New Zealand company that also wanted to be his agent. But he had told them he was already dealing with someone else. He asked me if I knew the company: Fletcher Challenge. Yes, I said, I did. I did not mention that it was New Zealand's biggest company and, apart from the Dairy Board, its only multinational! Score one for the little people.

The next thing, Richard was paying for an all-expenses trip to the UK for six weeks; in the English summer! I studied his operation in London and Peterborough, signed a heads of agreement, and returned to New Zealand all fired up and raring to go with a copy of the Micronet database.

"Why are you doing all this," I asked Richard as we signed the agreement, for an astonishing nothing (just a small share of any turnover later on).

"Someone did it for me," he replied. He was only about thirty, a millionaire, with a mansion north of London in a village called Shenley, near St Albans (Beebug's home; not that I knew that then). He put me up at his place for the first few days, as I rested after the 29-hour flight (I was in the Blue Room). Then I attended a Micronet meeting and a joint-venture meeting with BT in London (BT was extremely friendly and helpful), and after a few days in a hotel in Swiss Cottage, as I studied various things here and there, I went off to stay at the Bull in Peterborough and study Micronet's head office there (in Peterborough, not at the Bull).

While I was in London, Richard and his modem-making colleagues enabled me to acquire a BBC B at a very acceptable discount, so my initial desire had been fulfilled by my 'original' idea, but not quite as I had first imagined. It was not a case of a short trip across Auckland to pick up a rental machine; just a short trip to the other side of Planet E to buy one cheaply, with the New Zealand rights to the forerunner of the Internet as a side-dish.

The 'various things here and there' had included visiting the UK branch of Datapoint (the company that had invented the LAN, and was then its only significant manufacturer), and working out with its help how to set up the basic software structure for my planned host system. I also paid a visit to the HQ of the Nottingham Building Society to see how it used videotex in its customer service.

I had decided to write my own host software after seeing the crude way in which videotex worked. It had an upside-down tree structure through which you could navigate only two ways: by going down the tree to the page you wanted, starting from a basic menu, or by going directly to the page if you knew its address. Addresses were less gobbledegookish than today's Internet URLs, but they were still random messes. And if you went down the tree, then wanted to investigate another path, you had to go all the way back up the one you were on before you could go down the other one. Hopping straight across was impossible unless you knew the address of the other page. There was no word-access, no search-engine, nothing but a very crude user-interface. Navigation was a giant pain.

The upkeep of the system was even worse. Virtually every page had to be created manually. They also had to be linked manually and removed manually (these were not hypertext links, just entries in the text of the addresses of other pages). When a page was removed, following any link pointing to it flung you to the videotex equivalent of the notorious 404 message; in other words, you were dumped at a page that told you the page you wanted did not exist. And because pages had to be removed manually, that meant that the system slowly became cluttered with piles of obsolete stuff. The Internet is the same.

So although British Telecom's system was considered the bee's knees, the poor insect was really a grounded, brain-damaged cripple. I saw that it was possible to do a lot better than that, if I chose a host machine not limited in the same way as BT's mainframe.

Datapoint

They asked two makers of integrated circuits to make their microprocessor chip: Texas Instruments and Intel. Texas Instruments made it, then decided not to continue with the making of computer chips (what vision!). Intel came in a year later with the chip, and in 1971 Datapoint launched the world's first desktop minicomputer. Datapoint remained one generation ahead of Intel until the 80286, then changed to Intel's development-stream. The fact that Intel did not originate the 80X86 series is why others, such as AMD and Cyrix, can make clones with impunity.

An amusing aside was that, when the Datapoint trio was designing that chip in Victor Poor's living-room, Schmidt was only a teenager, and it fell to him to write the code that checked parity, obviously an important consideration in a chip that was going to be communicating. But he could not write code that ran fast enough, so he put parity flags into the chip itself. When I interviewed him in the 286 days, he said, "Funny thing, that flag is still there!" You can imagine Intel's engineers religiously keeping the things, not knowing that they were only there to cover a software deficiency that by then would have been long redundant.

Datapoint carried on being innovative. It invented its own programming language, Databus, a high-level Assembler (later accepted into the ANSI stable) which was very easy to write in, largely self-documenting if used properly, and very good for building communications- and user-interfaces. It was also an interpreted language (decades before Java), which ran very quickly and compiled at astonishing speed. You may wonder why an interpreted language was compiled. Databus did both. Compilation produced a very concise code of tokens that were interpreted at run-time, which created a very fast compilation and the most efficient interpretation.

Datapoint's first machine had only 16K of memory, but could drive several terminals, talked several programming languages, had DAT drives and featured a built-in screen (then unheard of). We would call it a PC.

Later, in 1977, Datapoint launched the world's first LAN. Digital liked to pretend that Ethernet was first, but Datapoint was well ahead, and its ARC (Attached Resource Computer, later renamed ARCNet) had a better throughput than Ethernet even though it ran at a quarter the baud-rate. It was also much easier to run. It used ordinary IBM co-ax (or infra-red beams between buildings), new machines could be added without taking the system down, and it was both very simple and very robust.

I had worked with Datapoint in the late 1970s, as MIS manager at New Zealand's second-biggest power utility, so I knew it well and had learned to program it. At that point I had one of the few LANs in New Zealand, and the biggest: all five nodes.

Designing a scalable system

Datapoint offered another advantage to a videotex host system. When a program starts it must open all the files it needs to use, which can be a time-consuming and thus very unfriendly process. The ideal would be to have the files open all the time, but, because they have to be opened whenever a user connects, that is impossible. But Datapoint had developed an operating system called RMS (Resource Management System), which took the best from Unix and was designed specifically for a LAN.

One feature that had been added to the Databus language for RMS was pipes (portions of memory through which programs could talk to each other), and when I visited Datapoint in London, a systems engineer suggested that I could use them to eliminate file-openings. For every telephone line I could have two programs. The first in each pair would come up when the machine was booted, and would open all the files (about three dozen); the second would come up when the line was accessed by a user, and would talk to the first through two pipes (one in each direction). The user would therefore not have to wait for files to open, just for the microseconds it took for the pipes to connect.

I soon realised that that structure created another advantage, because as the load grew as the database and/or the userbase grew, the 'network' of software could be expanded. For example, you might, as I planned to do, start with a single Datapoint machine acting as both fileserver and client-server (Datapoint called them File Processor and Application Processor, and copyrighted the terms, so the rest of the world had to use others). I started with one disc drive and eight lines, so when the machine was booted it loaded and ran eight programs, called Answer programs, that opened and accessed all the files users needed. When a user called in, a program called a Master program would load and run for his line, which talked to its corresponding Answer program. The Answer programs were thus soft file-servers; the Master programs soft client-servers.

Then, when the database/userbase load grew, there would be two machines, a file-server and a client-server. The first would run the eight `file-server' programs (the eight Answers), the second would drive the lines via the 'client-server' programs (the Masters).

As the load grew still more, the file-server programs could be spread across even more file-server hardware, thus sharing the file-access load. The file-server programs could also be chopped up so that, instead of each one accessing every file, they would share the access load, dividing up the files between them; that is, there would be two or more Answer programs talking to each Master program. In short, I would be building a software LAN inside a hardware LAN: a very elegant, powerful, infinitely scalable system, ideal for the task of being a videotex host (and easy to change later to an Internet host, although that was then a distant unknown).

With the benefit of all that knowledge and experience of the capabilities of Datapoint, I decided that that was exactly what I needed, and that on it I would write for MicroNet New Zealand completely new host-software with a powerful, friendly, intelligent user-interface. I say 'MicroNet New Zealand', but by then I had realised that my system would be able to do a lot more than Micronet did in the UK, so I chose a name for the whole operation, and planned to make MicroNet New Zealand a subsidiary of it. I wanted a name of eight letters (partly to fit the limitations of software labels), one that would be instantly recognisable, that came with an established history so as to make it instantly acceptable to the great unwashed, and that sounded strong. So I called it Hercules.

The Rotaview function key strip and reference card

Herculean power

By 'every single page would be indexed' I mean that it would be impossible to get any page into the system without its being put into every index automatically. And every page removed would automatically be removed from every index. No 404s here.

Every page also had a simple address, and all addresses were of the same format: AAA.ANNNa (that is, three letters, a dot, a letter, three digits, and a letter). The AAA.ANNN was the Page ID, the letter on the end was the Frame ID. The dot, so far as the user was concerned, was just to make the address easier to read and remember, like the dash in a telephone number. Pages in videotex, like teletext, were not scrollable: you flicked through a series of frames (screenfuls). My system provided up to 26 frames per page (more than enough to weary a user), hence the lower-case letter on the end. That address structure provided for 456,519,024 pages (26^4×999), which I considered enough for New Zealand.

But if the world needed more, I could easily add another letter, to make AAAA.A123a, or two more, AAAA.AA123a, thus making room for 11 billion or 30 billion, if those were addresses of the same sort. But the prefix letters could be the addresses of subordinate ARCs, which would mean that each letter would stand for an additional 456,519,024 pages, and because, at first blush, there could be 256 linked ARCs, that would make a possible total of 116 billion.

There was no hypertext in my system, partly because videotex did not have it, partly because I never thought of it, and partly because there was no universal WIMP interface in those days, so it would have been impossible. But even if I had thought of it, my system had a better method of getting you pages of like content; and hypertext has the enormous drawback of being impossible to keep up to date. If you have a million pages that have a link to a page, and you remove that page, the million are instantly made out of date. They now all point to the fearsome 404. They must all be corrected manually. They will not be, of course, so the whole thing gets worse and worse and worse...

Hypertext is there to link pages by content. But if that is done automatically by the system, hypertext is unnecessary. In videotex, displays were 40 characters across by 25 down (Mode 7 in the BBC Micro). In my Hercules system, the first 24 lines displayed the frame of the page you were on, starting with the ID line at the top; the bottom line was used for commands. So there were 23 lines of text and graphics.

The commands entered at the bottom of any frame could be simple things, such as the directives for frames and pages listed in the following table:

But commands could also be words: command-words, or ordinary words, or combinations of command-words and ordinary words. Command-words could almost all be abbreviated to the first two letters. The most basic was MASTER, which took you to the master page. It displayed a simple, ten-line menu, and waited, cursor flashing, two characters to the right of the colon after SELECTION:

You can see that the two-letter mnemonics for command-abbreviations are in the menu. That made the whole interface more friendly, because the Master Page then doubles as the basic Help Page.

Telemail was the term I invented for what we now call email. But in those days there was no term. TP was the abbreviation for telesoftware (tele-programs), rather than TS, because TS was used for teleshopping and two-letter abbreviations of command-words had to be unique. PH, the Phonebook system, was associated with YT, Yellowtronic pages, the online equivalent of the Yellow Pages. Both were part of the Hercules system. The sad, peremptory fate of PH and YT will be told later.

The user-interface was of course handled by the Master program. It talked to its paired Answer program, which accessed the Hercules database three ways: by Direct Access, by ISAM (aka KSAM), and by Associative Index Method.

Direct Access went straight to a sector, so the software translated AAA.A123a to the sector of a file (that is, sector A123 of file AAA, which means sector A×1000+123, where A could be anything from 0 to Z, and A-z stood for 1-26) and shot out its disc-head to get a match in one hit. Thus the address of a page was actually a disc address, generated by the editor, so you could not foul up page-addresses, and they could never point to something that was not there, or a there that was not something, or a not that was neither here nor there.

ISAM was the normal access via an ISAM key-file. I had ISAM files for page-subjects, so you could search pages on subjects, or the first few characters of them.

AIM indexed the subjects and alternative subjects of every page (you might have Widget as the subject and gadget, thingummy and device as the alternatives), and the text of the page in three-line chunks. So it could randomly access every page, so long as it was given at least three characters that matched either subject or text.

When you entered a word, or words, the system started by looking at the first one to see if it was a command-word, such as MASTER, or an abbreviation of it, down to MA. If it was a command-word it then looked at the second word, if there was one, to see if it was a command-word too. If so, it looked at the third word to see if it was also a command-word. For instance, the first word might be SUBJECT, or SU, meaning that the user wanted to restrict the search to the subjects of pages. The second word might be NEWS, or NE, meaning that the user was only interested in news pages; the pages whose frames all had their top line, the ID line, displayed in green on black (green is fresh, and news should be). More on third command-words later.

So you might enter, say, SUBJECT NEWS SPACE SHUTTLE COLUMBIA (I am using upper case here to make clear what the command is, not because it was needed). Hercules would then rush off and fetch all the pages whose subject was 'space shuttle Columbia' or whose subject began with those words. When it exhausted those, it would look for pages whose subject included 'space' and 'shuttle' and 'Columbia'. In both cases it restricted its search to news pages.

The possibilities for two particularly useful command-words, NEW and UPDATE, arose from another feature in the Hercules system (far more features than can be detailed here); one that removed a major videotex bugbear. It also made possible other features that are still not in the Internet. I mentioned earlier the way pages get out of date and have to be removed manually. So in Hercules every page could be entered with an expiry date. On that day it would automatically delete itself from the database and all indices. Pages could also be entered with a live date, so they would not be accessible until then. That was particularly useful for advertisements, because they could go live on a Tuesday and die on the following Friday, and be prepared weeks in advance. Pages also had add/change dates, so the system knew when a page had been created and last altered. That meant that Hercules could help you keep track of what had been added or changed in a given subject since you last called, or last asked.

For if you entered, say, NEW WIDGETS, Hercules would fetch every matching page that had been added or changed since you last called. If you entered UPDATE WIDGETS (or UP WIDGETS), it would limit itself to what had been added or changed since you last asked. You could, of course, use permutations and combinations of words, such as SU NEW ADV WIDGET, which would fetch every page with WIDGET in its subject that had been added or changed since you last called, but restricting itself to advertisement pages.

Beat that, Google! Not that Google could beat any of this. Google has indexed, after the fact, only a fraction of the pages in the Internet, via a crawler. Hercules indexed, at the time of entry, every single page. It ran in real time. It knew nothing about crawling.

The result was that Hercules had the appearance of artificial intelligence. It was uncanny. Once, on whim, I entered "what do I do about drugs abuse" (no question-marks were needed in such commands; if they were used, they were seen as wild-cards). A split second later the first page that popped up was headed "Don't Panic" in large letters. I was stunned. It took me about ten minutes to figure out how my own software had done that. It turned out to be very simple. First it had tried to access, using ISAM, a page with that string of characters as its subject, or whose subject began with it. When that failed it went through the string, discarded any word with fewer than three characters (AIM's lower threshold), and randomly searched the subjects with what was left. So it had searched for subjects and/or alternative subjects that contained 'what', 'about', 'drugs' and 'abuse'. There were pages about drugs abuse, which of course I knew, having put them in, and the first one it came to was headed "Don't Panic".

Another very powerful command was INDEX, or just IN. If you entered that, you got the index to the entire database, line by line, in alphabetical order of subject, assembled in real time: as up to date as your request. Each line was numbered, so you could jump to the page by entering that number. In other words, the index was a real-time, intelligent menu, assembled on request. If you entered INDEX WIDGET, you got an index only of pages that contained 'widget' in their subjects or alternative subjects. You could also enter INDEX NEW WIDGET, and get only the index for pages about widgets that had been added or changed since you last called. Or INDEX UPDATE WIDGET to get the ones since you last asked. Or INDEX UPDATE ADVERT WIDGET to restrict the display just to advertisements. And so on. You can see why hypertext was not needed, even if it had been invented. It is a very crude, manual way of linking pages; the Hercules system was entirely automated.

If you entered just WIDGET, Hercules would first fetch all the pages whose subject started with that word, then all those whose subject contained it, then all those whose text contained it. It kept going as long as you kept pressing <Return>, until it ran out of matches. But you could not only restrict its search, as described, to pages whose subjects matched; you could also restrict it to the text, by entering TEXT WIDGET (or TE WIDGET). Again, you could enter TEXT NEW ADVERT WIDGET to restrict it to the text of pages that contained the word 'widget', and were advertisements, and had been added or changed since you last called.

The NEW and UPDATE command-words could of course be used anywhere, even in Phonebook, the electronic telephone directory, and its accompanying Yellowtronic Pages.

I mentioned the ID colours for news. There were ten colour-combinations used for ID lines, to help navigation and make the user-interface easier. Apart from news (green on black) there was also telesoftware (red on black), teleshopping (yellow on blue), yellowtronic pages (yellow on black), service providers (blue on cyan), advertisements (magenta on black), information/general (cyan on black), classifieds (white on red), comments/letters (white on magenta), and education (white on black; chalk on the board: what else?). The same colours were used in the lines of the dynamic index, so users knew at a glance not only what the subjects were but also their category. Colour-coding, if used in system-design, has to be limited and easy to follow, which means that you must limit the number. No-one can keep track of dozens, but ten is easy to follow.

But the colours were not there just as a visual aid for users. They were also part of the index keys, so that the system could search in just one category.

The Labours of Hercules

When I came to create frames and pages, that was done in Rotaview on the BBC. The results were fired down the line to the Datapoint editor that I had written, which inserted it into the database. The latter added the page and frame subjects and IDs, keys, etc. and inserted the whole shebang into the right places in the Datapoint files and indices.

(Strenuous attempts were made for this 2003 article to resurrect some of the frames from twenty years before, but the Rotaview software refused to run under BeebIt's emulation in a way that would enable me to grab a single frame. I could get as far only as the first level of menus.)

Love's labours lost

Ulp! I had added to the mortgage on my house to finance my venture, and I had a superb product. But without a modem a computer cannot talk down a phone line, so customers were going to be non-existent. So after getting multiple variations on no from the NZPO's telecommunications hierarchy I went to war. My local MP happened to be the Minister of Justice. I sought his help, and, at last, in mid-1984, I had a meeting in the Beehive (Enzed's ministerial accommodation) with the Postmaster General, my MP and the NZPO's Director of Telecommunications Marketing. An undertaking was given to create a type-approval process. Afterwards, the aforesaid Director said to me: "I did not realise how important modems were to you." I was stunned. What do you say to a Director of Telecommunications Marketing who does not know that computers need modems to talk down telephone lines?

But there is many a slip between cup and lip. The NZPO, for some strange reason, found unaccountable delays, difficulties, problems et hoc genus omne between it and its goal of issuing a type-approval document. These things are so hard! Finally, after some 'slippage in the printing' the precious document was finalised and released. I recognised it immediately. It was a copy of the one created by British Telecom, which BT had given me a copy of when I was in London over a year earlier.

"Slippage in the printing!" What that really was was pure, unadulterated, 24-carat Yes, Minister. You, you little squirt, may have the Beehive and the Minister of Justice, and right, and truth, and a good idea on your side. But we have the keys to the copying machine. So there!

While all this was going on I was writing my host software. Most was done in 1984. So I also offered the NZPO my electronic phonebook and Yellow Pages free, in return for a mutually-acceptable profit-sharing arrangement, They did not even bother to reply to the letter. Vision? They were the same people who, when they were told that my system had full text-access, asked, "Why would you want that?" Insight!

I used to liken the attitude of the NZPO brigade to that of wild animals who mark out their territory with heaps of dung. If you come over the line they tear your throat out.

I also wrote, and had printed umpty copies of, an A5 90-page userbook (friendlier than 'user manual'). Again I used View. There was no DTP in those days. Or ArtWorks. It was all pretty basic. The book was printed on my daisy-wheel printer, pasted up and shot to film from there. The artwork on the cover was a mixture of pen and ink, Letraset, and a bit of photographic tweaking by the print-house. But it looked OK; as OK as anything could look given the technology, as the following images show (in greyscale, because dark blue and silver do not scan well).

Front and back covers of the Hercules User Book

The userbook was made A5 for ease of handling, and the inside covers had summaries of the Hercules Commands: frame and page directives on the inside front, and word-commands on the inside back. The MicroNet New Zealand logo is shown below, incorporating the stars from the New Zealand flag, which represent the Southern Cross (stars are also a good symbol for a network). It was redone in ArtWorks for this article by tracing over a scanned original; days of work was redone in less than an hour...

As the Yellowtronic Pages, Money Transfer etc. indicate, I had grander plans than just running a MicroNet New Zealand, which, as I said, is why I called it Hercules. The company was called Hercules Videotex Corporation. Its logo, shown on the user book's front and back covers, is somewhat blocky so that videotex's crude graphics could make a reasonable fist of it when rendering its dark blue and silver in blue and cyan. As the front cover shows, I billed Hercules, accurately, as "the most powerful and intelligent videotex system in the world." There was certainly nothing else with anything like its features, its power, its ease of use. And its simulated artificial intelligence was planets ahead, which is why the the slogan for the MicroNet New Zealand subsidiary was "For the microcomputer, nothing better has ever happened."

But as time went on and on, and was gobbled up by the NZPO's 'slippage' and other shenannigans, my original agreement with Richard Hease expired. He was still not getting a bean in return, because MicroNet New Zealand was not yet running, but he very generously extended it. But then in 1985 he moved on to other things. The new management said that it would extend the agreement... for £100,000! Needless to say, I did not sign, so lost the rights to use their database, which by then was the only thing of value in the agreement. But that no longer mattered much, because I had a growing database of my own, and I could still buy the modems and micro software and hardware from their UK maker.

Then my agreement with Datapoint New Zealand for a free minicomputer hit a snag, because the manager with whom I had made the agreement moved on to other things. The new management said that it would extend the agreement... for $1500 a month, to maintain their own machine!

But I was ready to launch. I had had extra telephone lines installed in my house for Hercules: a 50-pair cable for expansion, with eight connected to the Datapoint. So against a foreboding financial background, having used every mortgaged cent, Hercules Videotex at last went live in late 1985. I knew it was forlorn hope that I would get enough customers in three weeks, then every month, to be able to pay my overheads, especially that fearsome $1500. But I was determined to prove a point.

And I did. A customer who joined very early in the piece, who had done his masters thesis on videotex at a New Zealand university, excitedly emailed me: "It's a whole new world."

But, alas, it was a world that lasted only three weeks before I found myself unable to pay the $1500, or the rental on the lines. The NZPO pulled the plug; a tragic irony. All I had managed to achieve in two and a half years was to create the modem market in New Zealand... for everyone else.

I often think about those days, twenty years ago. I owned the forerunner of the Internet in New Zealand, I had a world-beating system, but it was all crushed flat by territorial bureaucrats devoid of knowledge, understanding, vision and humanity.

C'est la vie.