These introductory remarks are intended to raise briefly some of the main facts about the problem of storing the information required by an automatic computer in permanent or semi-permanent form. Some points will only be touched upon very lightly as it is known that they are more fully covered in later papers.
It is first necessary to clarify what is meant in this connection by the words permanent and semi-permanent. Briefly, they both describe systems which store information in a form that can exist without the necessity for the computer to be working. All high speed systems such as the ultrasonic delay tanks of the EDSAC or the cathode-ray-tube system used in the Manchester machine are thus excluded, as the information which they store disappears as soon as the power is switched off.
Such high speed systems form what may be called the working store, - sometimes known as inner memory, of a machine. Their function is to hold the information which is being used or which may be required for use during any particular stage, of a problem. A short time of access is the chief consideration here, whereas the capacity of the store need not be very great provided that it is backed up by what is known as an auxiliary store or outer memory.
This auxiliary store is needed to hold further information, not immediately required at any particular stage of a solution, but which must nevertheless be fairly readily available should it be required by the working store at a later stage. It must also be capable of accepting new information from the working store for use again later. It would not be practicable to transfer just isolated words in this way between working and auxiliary stores. In general, information would be recorded in the latter in blocks of suitable length and whole blocks transferred to or from the working store when necessary, only one instruction being used to effect the transfer of a complete block.
In this way an access time considerably in excess of that of the working store will suffice for an auxiliary store. Its capacity, however, must be much greater; certainly not less than about 105 words.
If the power supply of the machine fails, or the computer has to be switched off for any reason all information in the high speed store will be lost and will have to be built up again from the same sources as before. This might involve a considerable waste of time unless it could be transferred as a block or blocks directly from the auxiliary store. The auxiliary store must therefore be independent of power supply etc. i.e. it must be of a semi-permanent nature. It need not be absolutely permanent; at the completion of a problem the information recorded in it will generally become redundant and should be replaced by that relevant to the next problem.
There are two types of information, however, which do not become redundant at the end of a problem namely, sub-routines and tables of functions. A library of such information must gradually be built up and will continue to grow as new problems are solved. It is this library that requires permanent storage. The provision of such library facilities is often confused with that of an auxiliary store but in fact the two problems require quite separate consideration.
Before pursuing this point further it will perhaps be as well to say something about the actual physical systems which may possibly be used to provide permanent and semi-permanent storage. We may roughly subdivide them into:-
(a) Punched paper tape is the most commonly used medium at the moment. It is admittedly rather slow, being limited by the fact that it is a mechanical system, although the speed used for teleprinter purposes could probably be improved by various means by a factor of about 10. It is easily handled and stored and all the necessary editing equipment is either readily available or easily constructed and is not unduly complicated.
As tapes are easily copied they can be relied upon as a reasonably permanent library, so that in the EDSAC, for instance, paper tape provides not only the Input system, but also a library of sub-routines.
Similar remarks apply to punched card systems although they may be rather more expensive to run than those employing paper tape.
(b) Photographic and Photoelectric Methods.
The basis of a photographic method of storage is the translation of 1's and 0's into light and dark patches on a photographic film, glass plate or some other suitable medium. The digits are recovered by running the film so that the light and dark patches control the intensity of a beam of light falling on a photoelectric cell. This is completely analogous to sound-on-film recording in the cinema except that only two states of the film (all black and all white) are required.
A great increase in speed may be effected, however, by recording digits on film in a two dimensional raster. The relevant portion of the raster may then be scanned by a spot of light from a cathode ray tube, the information recorded on the film thereby controlling the intensity of that light which falls on the photo-cell. In this way continuous movement of the film is not necessary when reading a block of information. Speed of reading is therefore not limited by mechanical considerations.
The defect of such a system arises from the fact that photographic film requires processing before information recorded on it can be used. This makes it unsuitable for an auxiliary store where information may need to be written from the high-speed store and transferred back there again after a very short interval.
Photographic methods do, however, open up great possibilities for the provision of library facilities where rapid processing is not essential. Such methods are particularly attractive for the storage of tables of functions. Sub-routines can perhaps be dealt with more efficiently by slightly different means, about which more will be said later.
There is one other method, closely allied to the above which might offer a solution to the problem of providing auxiliary storage: that is the Phillips-Miller tape system.
This system records information mechanically by cutting away the black layer on a transparent base. The playing back can be accomplished as in ordinary photographic methods but the advantage here is that no processing whatsoever is required; information can be read back as soon as it has been recorded.
Recording is mechanical of course and its speed is therefore limited, but it is believed to be possible to run at a speed or about 16 feet per second which would provide storage for over 300 numbers per foot of tape. Several parallel tracks on one tape also seem feasible. Thus Phillips-Miller tape would seem to offer similar facilities to magnetic tape, which will be considered in a moment, except that new information cannot be written on top-of old on the same piece of tape. This is a disadvantage from one point of view but nevertheless leads to a more permanent system and removes the danger of accidentally losing valuable information.
(c) Magnetic Methods.
The principles of magnetic storage systems are discussed in the following paper by Mr. G.E. Thomas. As explained there, three physical systems are possible: wire, tape, or rotating drum. In each system it is possible to record between 50 and 150 binary digits per inch - i.e. up to about 45 words per foot.
If a magnetic wire system is chosen it is possible to work with a reel of up to 50,000 feet in length. This would hold over 106 words, but at the required speed of 20 feet per sec, it would take 40 minutes to run from one end to the other. A wire system therefore amply fulfills the requirements of capacity for auxiliary store but may have an undesirably long access time.
Plastic tape impregnated with iron oxide is another possibility. Here the length of one reel is limited to about 1500 feet owing to the fact that tape can only be stacked vertically and not laterally. The resulting capacity (about 5 x 1040 words) is comparatively small but can be improved by the use of several parallel tracks recorded on the one tape. One of the tracks may be used to provide correctly spaced timing or marker pulses and these in certain circumstances make the problem of control somewhat easier.
A third possibility is to record a large number of parallel tracks around the periphery of a rapidly rotating drum a few inches in diameter. Full details of two such systems now in use are described in the papers by Mr. Thomas and Mr. Tutchings. Using such systems an access time as low as 30 milliseconds may be achieved. This is of a very different order of magnitude from anything possible with tape or wire. Thus the magnetic drum seems to occupy rather a unique position somewhere between a high-speed working store and an auxiliary store. In fact in different machines it is being used in both capacities. Certainly it seems that by its use the size of the true high-speed store may be cut down considerably. Whether a further auxiliary store such as wire or tape would also be required is doubtful. It seems that adequate capacity could be obtained by the use of two or more drums in parallel.
Thus we have three possible magnetic methods of producing an auxiliary store. For a permanent library, however, certain other considerations arise. A library must not only be of large capacity, it must be capable of indefinite extension. Further, the information it contains must really be permanent; there must be no danger of its being accidentally lost.
This last condition is not satisfied by any magnetic system because the danger of accidental erasure or partial erasure by stray fields is always present. This is particularly true if facilities are provided for the machine to write information into the library in the same way that it would write information into its auxiliary store. In fact such facilities could not be tolerated if the library was to be relied upon as really permanent. Accidental writing of new information on top of old, whether by an error on the part of the programmer or due to a fault in the machine, would result in the complete loss of the old information. If this were information which itself had been written into the library by the machine i.e. information which did not exist on paper or in any other form, the situation could be really serious.
Thus it must not be possible for the machine to alter the contents of the library. The library would have to be prepared by some special means external to the machine. The machine, however, could be used as part of these special means provided it were in a special setup quite separate from the solution of any problem.
The condition of sufficient capacity and continuous extension could be met by the provision, as required, of extra reels of wire or tape or of extra drums, although it would seem that the number of drums eventually required might be prohibitive. Only the magnetic drum, however, is really capable of providing the necessary accessibility. The information required by any one problem might well be scattered very widely in the library in the case of tape or wire, and the resultant wait of several minutes for successive sub-routines to be assembled into the auxiliary store would be intolerable. The only solution would be to avoid wasting machine time on this operation by providing the necessary editing facilities outside the machine.
Unfortunately, such editing of magnetic wire etc. is not easy and requires quite complicated control equipment. Such equipment in fact bears not a little resemblance to the actual control equipment required for a serial machine.
All these considerations make a magnetic permanent store far from satisfactory. If a magnetic system has to be used to store library information its contents should be checked as frequently as possible and, moreover, preserved in some alternative form, e.g. paper tape.
This paper is not intended to go into any detail about Input systems, but a few remarks might not be out of place.
Input is essentially a slow process. Ultimately its speed is limited to the speed at which a human operator can operate a keyboard. Possibly the keyboard may impress the input information onto magnetic wire, etc., at slow speed the wire later feeding the information into the machine at a much greater speed, (it may be noted that the correct assembly of information during Input usually demands some computation-decimal-binary conversion for example - and the time taken for this computation places an upper limit on the speed of final input) but this is really irrelevant - the manipulation of the keyboard is the fundamental operation.
Moreover, this operation is one in which mistakes most easily arise and very comprehensive checks are necessary. Punched tape forms an admirable way of carrying out such checks and it would thus appear that punched tape should form an intermediate stage in the process of Input at least for some considerable time. Therefore there are strong arguments for having the basic library stored on punched tapes also. They are easily handled and from them sub-routines can readily be incorporated into programme tapes by means of a combined keyboard perforator/reperforator.
Finally it may be of interest to mention a suggestion which has been made in connection with the EDSAC. It is intended to store the sub-routines etc., on paper tape as usual. These will be fed into the high-speed store by the normal Input and will then be transferred in blocks to the auxiliary (magnetic wire) store.
This auxiliary store will be of fairly large capacity and will be subdivided into a small number of sections. One section will in general be allocated to a particular user who will be allowed to keep it for his exclusive use for as long as his problem requires. The machine is liable to be dealing with several problems during any period of say a week, each being tackled in several sections. Thus it will be possible for the different users to use the auxiliary store almost as a library in that they need only use tape to provide sub-routines at the beginning of the problem.
Accidents of course, are not ruled out. Carelessness on the part of any of the users, or faults in the machine could easily wipe out information which was still required. On the whole this should not happen very frequently, however, and when it does the situation is not irretrievable as the information, will still be available on paper tape. Any information generated during one part of a problem and which may be required at a later stage should, of course, be printed before the machine is temporarily turned over to another user.
