Appendix A. Analytical Machine Employment (1952)

This is a partial transcription of the Armed Forces Security Agency (AFSA) report dated 15 August 1952. The document was classified TOP SECRET codeword CANOE, Security Information, and declassified and approved for release 23 January 2014. I transcribed paragraphs 1-12 of the main document as being relevant to modern High-Performance Computing tradecraft. (William F. Friedman, "Report by the Inspector to the Director on Analytical Machine Employment, Dated 15 August 1952" August 15, 1952. The report's author is E.S.L. Goodwin, Captain, U.S. Navy, Inspector. It is part of the NSA's Friedman archives.)

Purpose of Analytical Machines

A force of people is the most flexible and versatile of analytical agents. It follows that an analytical machine, in order to be worthy of consideration for our processes, must have some advantage over a force of people that outweighs its disadvantage of inferior flexibility and versatility.

This advantage, it happens, is always one simply of speed. Every machine in our list is able to perform one or more processes faster than they could be performed by a force of people equal in size to the force required to operate and maintain the machine.

This gain in time benefits operations in two ways:

  1. It saves man-hours in certain operations which would be performed anyway (at least in part), with or without machinery, and thus permits machines to replace men, with gains in effectiveness and economy.
  2. It permits certain operations to be performed which, without the machine, would never be undertaken at all, even in part, in view of prohibitive or fantastic requirements in man-hours.

Logically, the second is no more than an extension of the first, but practically the split suggests a categorization which has some application to this analysis; that is, categorization by potentiality.

Machines according to Potentiality

By this criterion the machines fall into two categories:

  1. Labor-savers and extenders. Machines which replace men for operations which would be undertaken, at least in part, even without them.
  2. Revolutionizers. Machines which make possible attacks which could not be undertaken without them.
  • Revolutionizers now available: IBM, ABNER, ATLAS I, (GOLDBERG and CONNIE are borderline cases).
  • Revolutionizers projected: NOMAD, ATLAS II, (AFSAF 1-1 is a borderline case).
  • Labor-savers and extenders: All others, plus a large part of IBM.

There may be some quarrel with calling the medium and long cycle solution selectors "labor-savers" rather than "revolutionizers" on the ground that operational exploitations of certain systems would not be possible without them. While this is undeniable, it contains qualifications which make its whole truth quantitative rather than qualitative.

In any case, as will clearly emerge, counting these solution selectors as mere labor-extenders does no violence to the principles of their sound employment, and they are therefore, perhaps arbitrarily, so counted for the purposes of this analysis.

Employment Related to Potentiality

The categorization by potentiality leads directly to a philosophy of employment. The labor-savers and extenders should obviously be employed to perform every operation where they stand to save man-hours and speed up AFSA's results; the revolutionizers should be employed to the limit on operations which could not be attempted without them.

At first glance these two requirements look like slightly quibbled versions of the same thought, but actually they are far apart, in emphasis, in the two approaches to the machine employment problem which they generate.

The problem of making the labor-savers carry their weight is largely the straight-forward management problem of insuring, day by day and hour by hour, that operations are being performed as effectively and efficiently as possible.

If the demand for the talents of a certain labor-saving machine is such that it is active 20 hours per week, and it is determined that there are no residual tasks being performed by hand or by other machinery that could be done better by this machine, and it has no backlog, then its employment leaves nothing to be desired. It spends 20 hours each week saving labor and speeding results, and there is nothing more for it to do; its idle periods represent no loss and give no basis for reproach.

(I disregard here the argument that the space it stands idle in costs money. While this is true, the proper analysis of all the factors brings in the formula for the equivalence of man-hours and square-feet, involving concepts completely beyond the comprehension of all except those who are at home in the more stratospheric levels of Advanced Management.)

Thus SKATEs and DEMONs, of whatever vintage, have done their stints and may rest when there is no more of their kind of work to do. Whether there is, in fact, no more of their kind of work to do is a matter for 02's technical people to determine, as a computing part of their normal jobs of running the shop.

(Of course an expanded force of analysts might make more work for them to do. But the best conceivable reason for expanding the force would be to produce more for the consumers, and the desirability of this, from this angle, is beyond the scope of this study. The question of the desirability of expanding the analysis force solely to keep the DEMON from being idle in valuable floor space involves the formula for the equivalence of man-hours and square-feet, as mentioned in the preceding footnote, q.v.)

With the revolutionizers however, there seems to be need for an entirely different approach. If we have a machine that makes it possible to undertake analytical attacks that we could not undertake, even partially, without it, we would seem to be slighting our mission if we allow it to spend any significant time idle or performing labor-saving operations. If it has time available for a labor-saving operation, it should be so employed, but the moment this happens it should be a signal for the best brains to go into a huddle and devise some revolutionary employment to take over the time involved.

IBM, because of its nature, is an obvious exception to the principle that the revolutionizers should ideally not be employed in labor-saving. IBM is indispensable for labor-saving, and a large part of the IBM installation must always be available for labor-saving and extending operations. What this principle means with regard to IBM is that IBM's capacity for "revolutionary" application should be utilized; whether by extended shift operation or by procurement of additional machines is a point for determination based on logistic considerations.

Maximum employment of the labor savers involves simply good AFSA-02 management in the ordinary sense; full time of the revolutionizers, however, involves something on an altogether different plane, inventiveness and scientific imagination and analytical competence of the highest order. And the two require approaches from two different starting points; in the first case, the approach is "Which of these jobs can be done better by some machine?"; in the second, it should be "What can we get this machine to do?"

Present Employment Versus Reserve Capacity

It has been a matter of policy in this Agency that machine operations should be generally confined to one shift, with a view to maintaining in ready reserve the 66% of total capacity represented by the two idle shifts. Lest this policy be regarded as in conflict with the foregoing, let it be clarified here.

Surely it is not the real intent of the policy that we actually refrain from necessary or desirable machine employment to create an artificial or illusory reserve! Surely it is its real intent that we should procure and install, possibly not three times as much machinery as we need, but at least more than we need! No other interpretation makes sense, and insofar as this one does is the policy a valid working doctrine.

Employment of the Revolutionizers

ATLAS has been manned during six day-shifts, five evening-shifts, and five mid-shifts per week. This 128 hours per week is 76% of total available time (168 hours per week).

During the first three months of calendar 1952, because of light usage in the mid-shift, ATLAS was in actual use or under maintenance or alteration only 52% of total available time. This was increased to 72% during the second three months of the year, very nearly full utilization during manned shifts. A further breakdown of ATLAS's employment follows (I abbreviated "hours" to "hrs" for space):

Jan, Feb, Mar Apr, May, Jun
a Total hrs in three month period 2184 2184
b Total hrs manned during three month period 1664 1664
c Total hrs on "revolutionizer" jobs 774.2 1099.7
d Total hrs in maintenance, alteration, testing, or training 215.1 266.9
e Total hrs on labor-saving and extending jobs 144.6 201.6
f Total productive time (c + e) 918.8 1301.3
g Total non-idle time (c + d + e) 1133.9 1568.2
h Total idle time (a minus g) 1050.1 615.8
i Percent of total productive time as "revolutionizer" 84% 84%
j Percent of total productive time as labor-saver and extender 16% 16%
k Percent of total non-idle time as labor-saver and extender 13% 13%
l Percent of total available time as labor-saver and extender or idle 55% 38%

The last three items (j, k, and l) are the ones which, under the philosophy of paragraph 3 [section "Employment Related to Potentiality" above], should be minimized. It will be seen that, while idle time has been appreciably cut down in the second quarter, usage as a labor-saver and extender has remained proportionally constant.

Of the other revolutionizers, IBM is used in this role practically not at all (see para. 6, following [section "Employment of the Labor-Savers" below]), ABNER has only recently been installed and put through its initial operational tests, and CONNIE has been used 86% of only one shift exclusively in an essentially labor-extending role in the solution of PRDAW (French baud scrambler) settings. The other borderline case, GOLDBERG, likewise used in only one shift, is discussed in paragraph 6, following.

Employment of the Labor-Savers

In general, and with specific reservations, it may be said that the labor-savers are doing all of the big jobs and most of the small ones that the analysts ask of them. The specific reservations are as follows.

IBM

By far the greater part of IBM employment is for labor-saving and extension. The IBM complex has been working at about full capacity during one shift and half-capacity during a second (punches have worked at only a little over one-third capacity in the second shift, which indicates that preparation is not the limiting factor).

There is widespread complaint, in part, I believe, justified, that IBM results are slow, that much of the work is not sufficiently checked for accuracy, and that too much has had to be redone in whole or in part because of busts which should have been caught on the floor at the machines but which instead, are allowed to carry through to the end of the process, accumulating error to the final vitiation of results.

ROBIN

The ROBIN battery has been giving about 60% usage in one shift and 50% in a second. At the same time, the work it was provided for and is doing is not being pushed to completion with anything like the speed which, in the light of our mission, we should press for. As in the case of IBM, the difficulty here seems to be in operation of the machines themselves, not in preparation.

GOLDBERG

This operator has been used during one full shift of a 5-day week. Between 25% and 50% of this usage has been on selector jobs which could have been run, with less convenience and more delay (but necessarily less efficiency), on other machines.

Miscellaneous Small Operations

There are numbers of small and piece-meal machine jobs, requiring from a few minutes to five or six hours each, which do not get done and are not even asked for because of the complications inherent in the centralization of machine facilities. The analysts either do these by hand or pass them over and go on to other things or alternative attacks.

Discussion

These points are discussed in detail in the following paragraphs. It should be mentioned in passing that SKATE II furnishes a fine example of a labor-saver which seems to be well employed. It has been used, in three apparently well packed shifts (but, let it be said, with a rather high maintenance time; recently about 19%), almost solely on jobs which it is well fitted to do.

IBM's Labor-Saving Employment

The IBM plant's troubles seem to be due to inadequate qualifications and potentialities of personnel. The intake at the bottom has been largely in the GS-2 (entering punch operator) grade which, as the lowest civil-service grade, is the sump of the personnel pool, containing, along with some people of excellent potential, many of the drifters and part time vagrants, and some people who have no natural potential to achieve higher qualifications.

The GS-2s thus provide thin pickings for GS-4, 5, and 6 operator and supervisor material. In addition, not only the intake grade of GS-2 but the entire grade structure of the Machine Division is too low, by at least one or two grades, to provide for the kind of both talent and incentive that AFSA (as distinguished from other Government agencies) needs in its Machine Division.

Apparently, too, the outside labor pool of GS-4s, 5, and 6s is inadequate, perhaps for the same reasons that exist within AFSA. Added to this lack of adequate sources of replenishment of good floor talent is the fact that most of the few employees with this talent are in the Machine Division's front-office Methods Branch (AFSA-221), presumably because they are the only people qualified to man that Branch. The bad effects of this whole situation are two-fold, as follows:

  1. The people on the floor are not adequately qualified to know, from the analytical standpoint, what they are doing, and so, being unable to program their own job, have to receive detailed instructions which they follow blindly. Errors in both programming and execution tend to go undetected, and this is aggravated by the fact that some of the floor personnel are unreliable in performance of even some of the checks that they are qualified and required to make. The floor personnel tend, thus, to be a group of disinterested automatons, instead of alert members of the analytical team (as they were in both Agencies during WWII).
  2. In view of the situation on the floor, the Methods Branch is forced to spend much of its time in the detailed programming of individual IBM jobs, much of which is, to an able machine man, a fairly routine operation. The Methods Branch thus has itself largely tied up in work which belongs at operating-section level, or which, if it must be done at branch level, should be confined to the machine operating branches.

The effect is cumulative. The able people of the Methods Branch are unable to devote proper attention to the primary job of a "Methods Branch", which is to dream up new and better uses for machines; in other words, they are too busy with programming to be able to devote adequate creative and constructive study to scheduling.

At the same time they are absent from the operating floor, where, among other things, they are needed to develop their own replacements and raise the level of floor talent. The net result is a self perpetuating stasis which requires some deliberate boot-strap lifting. The Machine Division is moving toward the first pull in this direction by a program of temporarily shifting Methods Branch people to the floor, where they are to train and indoctrinate the operating section people in the analytical and programming phases of the work, with a view to eventually shifting the duties involved to the machine sections, where they belong.

For this move to be fully effective it will have to be accompanied by a program of raising both the grade levels of machine operating section jobs and the potential qualifications level of Machine Division recruits. A first move in this direction is already underway, an arrangement reached late in June between the Personnel Division and AFSA-02 whereby the standards for new punch operators was raised. Much more needs to be done in this direction.

It would appear to be not impossible to convince the Civil Service Commission that our machine operator needs are at least two grades higher than those of FSA and FHA; the ammunition for the attempt is clearly the understanding of the analytical process which we should demand in our operators, not excepting the punchers.

The ROBIN Job

The ROBIN program was pushed through as an interim measure on the basis that:

  1. No already available or imminently available solution was satisfactory, even as an interim one.
  2. The work could not wait for the ultimate solution (DELLA).

The second premise, imposed by AFSA-02, could not, of course, be challenged by anyone else. The first, however, was challenged by the then Head of AFSA-351 (Mr. Dumey) and appears to me, now reinforced by but not depending solely on some hindsight, to be vulnerable. The ROBINs, having cost close to a million dollars, are now under scrutiny as having produced a totally unexpected lack of results.

There is no reason yet to blame the ROBIN concept, and the million is by no means wasted, because the ROBINs will presumably always have uses, but I am convinced that a good deal less money spent improving AMBER would have permitted us (with the help of other existing machines) at least to check out the theory and approach of the ROBINs to find whether we really needed them.

There may be nothing wrong with the theory and approach, but the fact that we are even now checking these to see whether there is indicates the possibility that we crashed ahead without knowing too well where we were going. The ROBIN deal, incidentally, is the perfect answer to anyone who says that our approach to our problems is "*** and parsimonious". In my opinion this particular deal was extravagant.

ROBIN Utilization

Putting aside the history of the ROBINs, which cannot be altered and possibly should not if it could, we turn to the day by day uses of these machines. AFSA-02 could not wait for DELLA. Nevertheless, with DELLA now being installed, the ROBINs have been employed less than 50% capacity!

The reason is that there are not enough ROBIN operators to man the machines at full capacity 24 hours a day. The ROBIN problem thus boils down to one of personnel, requiring the same remedies that are needed for IBM; the Machine Division simply does not have enough people of high enough caliber.

ROBIN Preparation

The preparation of material for ROBIN illustrates a point which has a bearing on the Machine Division's personnel problems.

  1. The hard copy received from the field is first edited and collated by the analysts (AFSA-243) at Arlington Hall, then sent 8 miles to NAVSECSTA to be punched.
  2. The punching process includes the making of page proofs, which are checked by the punchers against the edited hard copies; tapes are corrected from this check and second page proofs are run.
  3. These are then sent 8 miles to Arlington Hall, where the analysts of 243 check them again.
  4. They are then sent 8 miles back to NAVSECSTA where the Machine Division's punchers make final corrections in the tapes, then run to make hard ("gray") tapes.
  5. The gray tapes are then sent 8 miles to Arlington Hall where they are run on the ROBINs.

It is not my purpose to emphasize the 32 miles of travel which this material undergoes; it is rather my purpose to point out the following:

  • Two distinct groups of people, 8 miles apart, engage in what is essentially a single routine process.
  • One of these groups has a grade level averaging GS-2/3, the other has a grade level averaging GS-4/5.
  • There is nothing more difficult about the routine operation of the GS-4/5 group than about that of the GS-2/3 group.

Aside from the argument that editing and collation belong, logically and economically, in the punching process (and are, in fact, often done there in other areas of our effort), it would seem to be possible to give the Machine Division's punchers higher responsibilities and higher grades, thus killing the following three birds:

  1. Helping to solve the Machine Division's personnel problems.
  2. Ridding what is ostensibly an analysts' group of a routine, mechanical operation.
  3. Saving total personnel, thus providing additional hands for, among other things, manning ROBINs.

GOLDBERG

This borderline revolutionizer's maintenance time runs extremely high; 35% to 50%. Disregarding this factor, which is conceivably capable of correction, its usage is of interest from two standpoints:

  1. It is used in only one shift, which suggests the possibility that a properly emancipated group of methods-inventors might be able to devise some uses to fill its spare time.
  2. With proper pressure being put on its schedule by the emancipated group of methods-inventors, it is conceivable that its present 25-50% employment as a selector would be shifted to its less versatile cousins, to permit its full-time use in its higher capability as an operator.

Some of GOLDBERG's less versatile cousins are now employed only part time (in one shift) or no time at all. Consideration of these, which are counters and the simpler kinds of selectors, and their possible employment leads directly into the next two paragraphs.

Miscellaneous Small Operations

Every analytical section in 02 has a constant need for small piecemeal jobs in which machines can minimize man-hours. Some of these, of a purely mathematical kind, are performed with the aid of the analysts' desk-top Friden calculators.

An analyst wishes to make a mathematical computation. If he has no Friden, he can do one of two things; work it out by hand or follow the S.O.P. to have the job done by the Machine Division on the IBM-604 (or O'MALLEY, if appropriate). If the job is a very large one, he will, of course, write it up and send it to the Machine Division, and will get his answer in not less than two days. If it is anything other than very large, and he wants the answer right away, to fit in with a current train of thought, he will probably make a stab at working it out by hand. Possession of the Friden means that he is assisted and speeded up in the operation; non-possession of it means that the job will not get done or will get done inefficiently.

The arithmetical job has been used purely as an example; there are many counting and selecting jobs to which the same arguments apply. Two of the basic chores of cryptanalysis, monographic and digraphic frequency counting, are important cases in point. An analyst may want counts taken on one or two messages rather than a thousand. The thousand, he would write up and send to the Machine Division; the one or two he would toss to a clerk to count manually, unless he has a machine counter handy and informally accessible.

Possibly the count of one or two messages convinces him that he wants the count on the entire thousand, in which case he will go through the formality of the work-request to AFSA-22 for the big job. But possibly, instead, the count of one or two convinces him that there's nothing in his hunch, so he abandons the idea of further counting.

And this last is important, for if he cannot conveniently check his idea on the small sample he may request the big job of AFSA-22 without a check (and get his negative result two days later after a considerable amount of effort by 22 instead of one hour later after an easy run on his own machine). Exactly similar considerations apply to sliding coincidence and differencing runs. On the whole, the lack of certain machines directly and informally available to the analysts leads to the following inefficiencies:

  1. Work which should be done is not done, or
  2. It is done by hand methods, or
  3. It is unnecessarily inflated to make it big and important enough to be put through S.O.P. for performance by the central machine pool; or it is ordered in toto even when a first piecemeal result might show the complete job to be unnecessary.

The principles involved here are recognized, and several sections have their small machine rooms, adjacent to their main areas. There is, however, a mild but constant pressure to minimize this dispersion and to prevent its further growth; there seems to be a tendency to accept as axiomatic the theory that if you have something which is a machine, and something called a Machine Division, the machine must obviously belong to the Machine Division. As a result, some sections do not have all the locally controlled machines they need, and some have none at all.