Re: Application design & architecture -- MIDRANGE-L

We got hijacked and taken down a blind alley.

Let's try and get back on track.

I asked - it seems eons ago -

"Can we design an architecture that will make us more efficient?"

I don't think that many people will regard present methods as efficient- just the best that they know.

There are many reasons for this, but I would like to concentrate here onjust one of these - programming.

Whatever the language or the architecture, programming is a mostinefficient process, as I said earlier, "the most inefficient processever devised by mankind in any field of human endeavour".

In order to explain how we might create a better moustrap, a littlecomputing history is relevant. I have tried to keep this as short aspossible.

The first computers were conceived and constructed to do high speedcalculations - they were built solely for computation. John von Neumannand his colleagues, working in 1944 on the Manhattan project to buildthe first atomic bomb, wanted a much quicker way of solving complexmathematical equations. Storing, manipulating and constantly updatingvast quantities of data was not the problem that von Neumann was tryingto solve.

Von Neumann, who first conceived the required logical structure of thecomputer, realised that his mathematical problems could be closelydefined but needed to be run repeatedly on many sets of data. He furtherconceived the concept of the stored program as a convenient way ofdefining each separate problem to the computer and allowing hiscolleagues to switch quickly between the different problems they weretrying to solve. It is unlikely that von Neumann considered that his"model" would be used for all computing work for the next 60 yearswithout any real change.The separation of the rules from the data may be appropriate formathematical and scientific problems but has added dramatically to thecomplexity of creating and maintaining business computer systems. Atthe same time, we may have ensured that we can never create a machinewith an ability that might be called "intelligent", whether or not thatis similar to the intelligence of the brain. We may call the presentartificial split of rules and data in computer systems "*Separatism*".It is clear that human intelligence is very dependent on our ability tostore, dynamically update, and navigate large numbers of connectionsbetween the items of data in our brain. This ability, which may beconsidered to be non-computational, can conveniently be described as"*Connectionism*".

If we are to move forward, the rules and processes must be defined, notin the separatist manner using programs and files, but in associationwith the data to which they apply, using connections. This means that,like the brain, our computer database must be able to determine whereand how information is stored without outside help (from programchanges). In other words, it must be capable of defining its ownstructure and allow continual changing of the connections as it acquiresnew data. All of this is way beyond the capability of RDMBSs but not sodifficult to implement if we create a universal data structure.

Applying the principle of connectionism to the building of computersystems can allow us to build systems with a speed and ease that isbeyond the wildest dreams of those using separatism as their model.Connectionist systems can be changed dynamically whilst they are in useand so continually be kept up to date with the world of their users.Since connectionist systems are defined using the language andterminology of their users, the process is one to which users arealready accustomed. Typing their business descriptions, definitions andrules into a connectionist system will be dramatically simpler to learnthan the present separatist processes of system design and programming.As with the information stored in their brain, users will not need tounderstand the complexity of the structure of their world, since theycan study and define it incrementally in small pieces and use theconnections to move from one part of the definitions to another. Monthsof detailed planning and specifying become a thing of the past.

Separatist methods produce static rigid solutions whilst theunstructured real-world problems of their users change dynamically.Static solutions are totally inappropriate for any business that istrying to get ahead or even to stay put. The effect of separatism is"*isolationism*". As John Donne (1571? - 1631) wrote in his ~Devotions~,"No man is an island, entire of it self".

I wrote the above many years ago in isolation (!!) in order to try andexplain the connectionist system that I had built and then found thatthe concept of connectionism had been invented by Dr. Vannevar Bush in1945. It may all seem very removed from the present day world ofcomputing but it is an extremely simple concept. It can be implementedin about 6Mb of RPG code (any flavour will do) and the same code is usedfor both development and also for operation. One of the additionalbenefits of connectionism is that it ticks many important boxes - e.g.security, performance, scaleability, documentation, standards, objectoriented, automatic re-use of definitions and code, contextual help,multi-lingual, etc. - and it allows sharing of definitions on differentsystems so that developers don't need to spend their time reinventingthe wheel. Rather than having open source programs, we can have an openbusiness model to which multiple users can contribute and which they canuse.

Connectionism actually achieves much of what SOA attempts to do,because, unlike SOA, connectionism is a technology that can beimplemented. If we.are to progress, we must stop using von Neumann'smodel for a purpose for which it was not invented - running a business.

Of course, my implementation of connectionism has not solved all theproblems of the world. I have done it as a one man band, in isolationand I trained at IBM as a salesman, not as a techie. Anyone else mightbe able to do a better implementation of the concept of connectionism.


Rob Dixon
www.erros.co.uk