The discussion was introduced by Dr. S.H.Hollingdale who described the R.A.E. Sequence Controlled Calculator as follows:-
Much of what I may call the bread and butter computing at an experimental establishment such as R.A.E. consists of the reduction and analysis of experimental results. This type of work leads to moderately long sequences, often including reading from tables of mathematical or empirical functions, with frequent repetition on successive sets of data.
A demand accordingly arose for an automatic calculating machine intermediate between the desk machine and the large electronic calculator. A relay machine which seeks to fill this need - known as the Sequence Controlled Calculator - has been designed and the prototype is now being built by Mr. Petherick of R.A.E., who is unfortunately unable to be present. Speed has been sacrificed to low cost and emphasis has been placed on ease of operation and resetting, and on simplicity of programming.
The machine handles 7-digit numbers expressed in the decimal system with a floating decimal point. A number N is represented as p × 10j where
1 ≤ p < 10 (to 7 figures) and -19 ≤ j ≤ 29
This range of j enables the sign and index of a number to be coded with 2 decimal digits, and as 8 figures are retained throughout to minimise round-off errors, a number within the machine therefore requires 10-decimal digits altogether. The floating decimal point, while somewhat complicating the equipment, was considered to be worthwhile in the interests of simplicity of programming.
A decimal digit in transit is represented by an equal number of pulses along 1 wire. The computer is parallel-wired; the several digits of a number not being combined into words along a single wire.
Power is taken-from the mains and the pulse emitter supplies 100 volt half wave rectified pulses with a repetition frequency of 50 per second.
Input of instructions, numerical data and tabulated function is by means of punched tape (in fact, a 4-hole code on uncoated 35 mm. cine film will be used).
The output is arranged to operate one or more I.B.M. electromatic typewriters for page printing.
There are 2 kinds of storage; a relatively small internal storage of 80 electromechanical registers and a much larger external storage on punched tape. The computer proper and the internal storage consist almost entirely of a pulsed network of simple P.O. relays and uniselector type switches, the circuits of which are changed when no current is flowing.
There are 8 built in tape readers and 10 sockets, into which can be plugged extra tape reading units, typewriters, or repunching units dealing with intermediate results that will be required later (the external storage). These 8 readers, 10 outlets, the Main Register and a special, zero source for clearance purposes, together with the 30 internal stores, give 100 possible sources and destinations, any of which can be specified in a coded instruction by 2 decimal digits.
The machine uses a 3 address code - 2 sources and 1 destination being usually specified. Each successive operation is allotted its own control instruction, which commences with a pair of digits specifying the operation, followed by a maximum of 3 other digit pairs specifying sources and destinations. There is thus the possibility of 99 different control instructions, and a considerable number have, in fact, been used in order to simplify and shorten the programming.
The input tapes are of several types; each calculation demands a master control tape, and also a display tape. When the print code appears at the end of an instruction, the display tape is consulted to give the page-positioning of the printed result. Other tapes provide numerical data (in decimal or angular measure), constants, parameters, and tabulated functions (other than sines and cosines, which are built in). There are also subsequence tapes whereby subroutines may be introduced under control of the master tape.
A point to be noted is that the control instructions (either on the master or subsequence tapes) are not transferred, to the stores, but are read sequentially from the control tapes.
Flexibility is provided by the use of a number of subsequences and the usual discrimination and modification operations are possible. Each subsequence tape may contain up to 99 subsequences, each being specified by a 2 digit number. While the internal stores are normally used to store numbers in standard form, the 10 digit positions can be split into 5 pairs, the position of the pair in the store being specified by a digit pair, 01 to 05. Sources and destinations can be defined by such digit pairs, with considerable increase in flexibility since digit pairs can be modified at will. The sources and destinations called for by a subsequence can also be modified in successive repetitions by digit pair specifications.
Each decimal digit retained in a store is memorised by 1 switch, which can be pulsed by current to its coil. An armature so attracted springs back to its dormant position after each current pulse, stepping a rotor through 1/10 of a revolution during its return. This selector is a simplified version of the standard G.P.O. uniselector.
If we number the 10 rotor states 0 to 9, then in state 9 a pair of monitor contacts is closed. From its zero position, a selector can be set with any digit 1 - 9 by an equal number of pulses to its coil. The digits can be read by feeding 10 pulses to the coil, and counting those, less 1, arriving after the contacts close; preceding pulses form the complement on 9. The selector may be cleared by supplying pulses only until one passes through the monitor contacts.
There are 2 main registers, but on]y one is used in addition and subtraction. It consists of a ring of 25 basic digit accumulators, each being a selector of the type just described. No operation requires more than 10, but the provision of 25, with shift facilities - which are of course necessary with a floating decimal point - spreads wear. To add two positive numbers, we may assume that 10 adjacent accumulators have been selected by the shift uniselectors. The 8 digits of a number are then set in parallel on accumulators 2 - 9 by feeding the necessary numbers of pulses. This takes place in the first 9 pulses of a cycle of 10. If another number is added similarly in a second cycle, any accumulator except the first may step past its ninth position. A pulse which does this emerges through the monitor contacts and steps a corresponding carry switch. Inter-digit carry is deferred until setting is complete and is effected simultaneously - using the principle explained by Prof. Hartree in his opening lecture - by the 10th pulse of the cycle.
Subtraction entails the addition of complements on 9 with the usual fugitive one on the right. End Carry from the left-most selected accumulator indicates either the sign of the answer or the required adjustment of the index to preserve standard form.
Multiplication requires 2 main registers (MRA and MRB) and 2 auxiliary registers (ARA and ARB), together with a built in multiplication table which is served by a pulsing unit.
The multiplicand is set on ARA and the multiplier on ARB, digits being set in parallel. Let αn be the nth digit of the multiplicand. Any digit β of the multiplier can be set on the Mult. Table in 1 cycle of the Pulsing Unit, by closing an appropriate relay in M.T. The Pulsing Unit is thus connected for one cycle to ARA in such a way that γ and δ pulses emerge along 2 output cables from ARA, where
αn × β = 10 γn + δn (n = 1 - 8)
These pulses proceed to MRB and MRA respectively.
Multiplication of 8 figure numbers can thus take place in 8 cycles of the pulsing unit, the units partial product being formed in MRA and the tens partial product in MRB. The final product is formed in MRA in a 9th cycle, when any necessary shift of the decimal point to preserve standard form is also arranged.
In division the dividend is set on MRB, the divisor on ARA and the quotient appears in MRA, the remainder appearing progressively in MRB. ARB is not used.
Square rooting which is built in, resembles division, a mock divisor being employed.
Prof. HARTREE pointed out that the system of multiplication described by Dr. HOLLINGDALE was almost a relay analogue of that used in the ENIAC except that it used a floating decimal point, whereas the ENIAC had a fixed one. He said that Miss BRITTEN had spoken of contact wear as though it was due to the contacts opening and closing when alive electrically, whereas he had always understood that relay contacts should only open or close when electrically dead. He was interested in the statement that the cost of valves was approximately equivalent to that of relays since it seemed to follow that valves provided extra speed for nothing.
Mr. KILBURN raised the question of relative reliability of relays and valves and Mr. REY asked if there could be any advantage in mixing relays and valves since relays were comparatively slow.
Dr. COOMBS said the G.P.O. had always been interested in the question of when to use relays and when to use valves, and in their relative reliability. The worst characteristic of valves was that the heaters developed discontinuities. Many valves were necessary in high speed computing machines and these consumed current whether they were being used or not.
The trouble with the standard P.O. relay was wear. The residual stud became battered after a time and required adjustment. Dust was a severe problem in high speed relays such as the Siemens type since it gave a transient fault which often disappeared before inspection and cleaning. The standard P.O. type 3,000 relay had twin contacts which greatly reduced the danger of intermittent faults due to dust. Relays should always be mounted, if possible, with the contacts in the vertical plane.
Relays were cheap when compared with the number of valves required for the same purpose and involved less auxiliary equipment. Relays needed only 50 volts which could be provided by battery or rectifier. His wartime experiences of mixed valve and relay operation had shown that valves were preferable for continuous use and relays better for occasional work.
Mr. I.J.GOQD said that valve heaters might be expected to improve as a result of research now in progress. Failures were reduced to a minimum if the heaters were never switched off.
Mr. WILKES said that the BTL and Harvard relay machines are designed on quite different principles. In the BTL machine no relay can operate until the previous operation in the sequence has taken place. The speed of the machine is thus governed by the speed at which the relays can operate, and the machine stops if one fails to operate. On the other hand, in the Harvard machine all relays are controlled by a set of motor driven contact breakers, which determine the speed at which the machine operates. This speed is so chosen that there is ample time for the operation of the relays. In the Harvard machine, the relays are not called on to break current, but in the BTL machine they are.
Dr. WALLMAN thought that the transistors now being developed would eventually replace valves for computing and similar purposes since they had no heaters and would probably be smaller as well as more economical.
Mr. K.E.MACHIN asked whether the Post Office favoured the use of relays in which some contacts operated before others. Dr. A.W.M.COOMBS said that the Post Office Research Department now worked on the principle that a relay was either on or off, and consequently did not favour make before break or anything fancy.
In answer to a question, Dr. COOMBS further said that hermetically sealed plug-in relays with mercury contacts had been tested in America but not used extensively, and Mr. WILKES recalled seeing one at the Physical Society's exhibition which however did not plug in.
