A Discussion of Proposals by Vendors for the Hybrid Simulation Equipment for the C-3 Project (LMSC)
A DISCUSSION OF PROPOSALS BY VENDORS FOR THE HYBRID
SIMULATION EQUIPMENT WHICH IS TO BE INSTALLED IN
BLDG. 161 FOR THE C-3 PROJECT
On March 25, 1966, J. E. Sherman, Manager of Analog and Hybrid Computing, made a presentation on the
selection of a vendor for the Hybrid Computing Center to some 40 IMSC people including representatives
from the C-3 Project Office, Information Processing, MSD Facilities, Procurement, and “/m@mic Analysis
and Simulation.
The presentation was in the form of a discussion accompanied by a set of charts. The charts plus
&@ transcription of the discussion follows. It must be remembered that this is a literal transcription
and is intelligible only in conjunction with the charts. It may be that the grammatical construction
of the accompanying text is sometimes inadequate because this is a literal transcription.
QAotee.,
E. Sherman, Manager “~~~
dog and Hybrid Computing
HYBRID COMPUTER
SYSTEM
Chart No. 1
The purpose of this meeting is to discuss how we selected a hybrid computer system for use on the C-3
Project. First, I would like to introduce to you the people who worked on the project. Tom Anderson,
Ken Bedient, Dee Foote, and Ralph Wheeler. These people have been working on this project since last
November. The first thing they did was to prepare a Request for Proposal which was sent to various
vendors. They have done an excellent job on this document. It is quite detailed and quite complete.
We have received a number of favorable comments from the vendors who have examined it.
As a sample of the kinds of replies we got, I've selected replies from three of the vendors and brought
those along. Electronic Associates, Inc., Astrodata, and Beckman. This gives you some idea of the
magnitude of the work of evaluating these proposals.
HYBRID SYSTEM
Chart No. 2
First, I think I should tell you what a hybrid computer is. A hybrid computer is composed of an analog
computer, an interface system, and a digital computer. What we have to do is to constrain these three
diverse pieces of equipment to work as a single system.
HYBRID SYSTEM CONFIGURATION
| CPU INTERFACE ANALOG
ANALOG
48K
COMMON PERIPHERALS
STORAGE
a ANALOG
CPU INTERFACE ANALOG
Chart No. 3
The particular hybrid system that we have conceived to handle the problem posed by the C3 Project
consists of two central processing units connected to a common core storage of approximately 44,000
words. They are connected to the normal digital peripheral equipment, two pieces of interface equip-
ment and four analog computers.
LMSC'S HYBRID COMPUTER SYSTEM
REQUEST FOR PROPOSAL
| BIDS RECEIVED
ASTRODATA INC (COMCOR)
APPLIED DYNAMICS INC.
BECKMAN INSTRUMENTS
CONTROL DATA CORP.
ELECTRONIC ASSOCIATES INC.
INTERNATIONAL BUSINESS
MACHINES
MILGO ELECTRONIC CORP
RAYTHEON COMPUTER
REEVES INSTRUMENT CO
SPERRY RAND CORP (UNIVAC)
NO BID
ADAGE INC
COMPUTER CONTROLS CO.
DIGITAL EQUIPMENT CORP
GPS. INSTRUMENT CO.
SCIENTIFIC DATA SYSTEMS
Chart No. 4&
We sent bids to 15 different companies. Of the 15 companies, five of them did not reply and
ten replied.
RESPONSES TO LMSCS RFP
ASTRODATA (COMCOR)___SEVERAL HYBRID SYSTEMS ___ MEETS ALL MAJOR REQMTS
APPLIED DYNAMICS AD 256 ANALOG ONLY
DID NOT MEET OUR REQMTS
BECKMAN 2200/$08 9300 HYBRID MEETS ALL MAJOR REQMTS
CONTROL DATA 3300 DIGITAL+ADCOM INTERFACE _ SAME OFFERED WITH ASTRODATA
ELECTRONIC ASSOCIATES___ 8900 HYBRID SYSTEM_______ MEETS ALL MAJOR REQMTS
IBM 360-44 DIGITAL ONLY DELIVERY TOO LONG
MILGO. 4100 ANALOG ONLY___—sSiéDID NOT MEET OUR REQMTS
RAYTHEON 520 DIGITAL ONLY_______ TOO SLOW FOR OUR REQMTS
REEVES 600/SDS 9300 HYBRID DID NOT MEET OUR REQMTS
UNIVAC 494 DIGITAL ONLY _____——dDELIVERY TOO LONG
Chart No. 5
Of the ten companies who replied, it was soon apparent that three of the bids were vastly superior to
the others. Those were the bids sent in by Astrodata, Beckman, and EAI. The other bids were simply
not responsive to our Request for Proposal in one or more major requirements. Applied Dynamics bid on
the analog portion of the system and they bid their standard Analog Computer, and did not really follow
the particular requirements which we put forth in our Request for Quotation.
Milgo, which is a new entry into the analog computer field, makes a very interesting analog computer,
but they felt they did not want to enter a serious bid at this time. They are in the process of
changing from a few specially built systems to a standard line of analog computers, and they are simply
not ready at this time. The Reeves Company, one of the old timers in the analog computer industry,
has come out with a brand new analog computer, and they did bid their computer in conjunction with SDS
digital computer, but unfortunately, the Reeves computer is just too new at this time and is not well
enough established. Further, they bid their standard computer system which again was not in line with
our requirements. Both IBM and Univac bid on the digital portion and both of them had rather interesting
digital computer systems -- particularly the IBM 360144 -- but unfortunately, the delivery date is much
too long. Raytheon bid on the digital computer portion, but the computer was too small and too slow to
meet our basic requirements. CDC bid a 3300 digital computer and they also bid an interface made by
their subsidiary, Adcomp. The 3300 digital computer is also included as a sub-bid in the Astrodata
system. We had previously evaluated the Adcomp interface, and it simply was not as good as the interface
system offered by the other people.
MAJOR ELEMENTS IN AN ANALOG SYSTEM
@ BASIC CONSOLE @ 30 LIMITERS
@EDVM @ COMPARATORS
Scorn © ELECTRONIC SW
@ ETC. @ RELAYS
@ 60 INTEGRATORS @ 240 POTENTIOMETERS
@ 60 SUMMERS @ 200 LOGIC UNITS
© 90 MULTIPLIERS @ 3 EIGHT CHANNEL RECORDERS
@ 6 RESOLVERS @ 3 X-Y PLOTTERS
@ 24 CARD SET FUNCTION GENERATORS
@ SHARED BY THE FOUR ANALOG COMPUTERS
@ TEST RACK
@|/0 SYSTEM
@ SPARE PARTS
Chart No. 6
Now, I want to point out to you the major elements in an analog computing system. There are four
of these analog systems in the overall hybrid computer system. Each of the four consists of a basic
console, that is the control console which contains a digital voltmeter, oscilloscope, a remote control
voltage divider, and other pleces of equipment to operate the system. It has, for example, the readout
selector switch and things of that nature. In active components, they have approximately 60 integrators,
60 summing amplifiers, yO multipliers, 6 resolvers, 24 function generators, 30 limiters, 240 potentiometers,
200 logic units, (3) B-channel strip chart recorders, and 3 x-y plotters. Now, contained in the non-
linear equipment, is another 300 inverting amplifiers. In addition we have an adequate quantity of
miscellaneous equipment such as comparators, electronic switches, relays, function switches, and things
of a like nature. The logic units are something that are fairly new to the analog computing field.
These are flip-flops, gates -- devices of this nature. Im addition to the equipment which is in each of
the four analog computers, there is another group of equipment which is shared by all of the four machines,
namely, the test rack for maintenance purposes, an I/O system which will be used when the digital computer,
which is part of the hybrid system is not available, and the spare parts necessary to maintain the system.
PERFORMANCE COMPARISON OF ANALOG SYSTEMS
ASTRODATA BECKMAN ELECTRONIC ASSOC.
Cl 5000 2200 8800
@ AMPLIFIER | 2 3
@ MULTIPLIER 2- 2 |
@ RESOLVER 2 3 |
@ CARDSET FUNCTION
GENERATOR | 2 |
@ LOGIC 2 3 —|
@ |/0 SYSTEM | 2 3
@ CROSSTALK | 2 |
@ DRIFT | 2 |
@ CONSTRUCTION 2 3 |
@ MAINTAINABILITY 2 3 |
MEAN 5 2.4 l4
Chart No. 7
Now for a performance comparison of the analog portion of the bids put out by the three major bidders.
This is really 4 subjective addition of a series of objective tests and examinations. The detailed
specifications are quite difficult to describe to a large group ~~ mainly, they are in the form of
graphs and plots -- so we have merely given a summation at this point. For example, when we evaluated
the amplifiers, we looked at such things as the total error curve as a function of frequency; the change
in amplitude as a function of frequency; the change in phase angle as a function of frequency; the noise
level; the output impedance, and various other characteristics and qualifications. The summation of all
these indicate to us that the amplifier offered by the Comcor subsidiary of Astrodata is the best of the
three; the Beckman amplifier being second best, and EAI coming out third best. Now this is rather
surprising because BAI is, after all, the leading manufacturer of analog computers, has been for some
time. They have recently redesigned their analog computing system. The 46800 is the latest development
in that line, and it has involved a complete redesign of the computer and all the components in it,
including the amplifier. According to the test data furnished to us by EAI, and supplemented by
additional measurements which we have made ourselves, they just seem to have made a biunder when they
designed the amplifier. Their total error at & cycles is .2 volts, Now that's more than double the
error at that frequency, or any nearby frequency, of any of the competitors. The output impedance is
much higher than that offered by its competitors. The frequency response is not as good. In talking
these matters over with EAI‘s technical pepresentative, we learned that they plan to redesign the amplifiers
some time this year, They know it isn’t quite right, and they are going to do something about it, but
we don't know at this time exactly what it is they are going to do or how well it's going to turn out.
However, in multipliers and most of the rest of the items, EAI comes out in first place. One exception
being the I/O syatem. Now the basic I/O system that EAI offers with their analog computer is very
good. It is a small digital computer. But, since we are buying this as part of a hybrid system, we
will have fairly large digital computers that will be available for the I/O work most of the time. How-
ever, when they are not available, we would like some kind of an automatic I/O system tc the computers.
However, it can be fairly simple -- a punch paper tape reader that will enable us to set potentiometers,
turn the computer on and off, and a few simple things like this. EAI just does not have anything adequate
in the small area, The only thing they can offer us is the small digital computer costing some forty
thousand dollars. Both of the other people are able to offer us adequate items for imuch less money. In
everything else, EAI comes out in first place. The Comecor equipment offered by Astrudata comes out either
tied with EAI or following them very closely in second place. Beckman comes out either second or third
in all regards. This is primarily because the Beckman computers have been designed a number of years ago
Chart No. 7 (con'td)
and are really obsolete when compared with the units offered by the other two people. Now, if we take
the mean of the subject ratings, we come up with a subjective overall performance index, and this shows
that EAI, if we can somehow overcome the major difficulties in the amplifier, is probably the best of
the analog computers. However, they are rather closely followed by Comcor/Astrodata <-- Beckman comes in
a fairly poor third.
INTERFACE FUNCTIONAL BLOCK DIAGRAM
JANALOG TO DIGITAL
CONVERTERS [~
DIGITAL TO ANALOG .
CONVERTERS
IGENERAL PURPOSE
LOGIC *|INTER-| | DISTRIBUTION
COMPUTER] PATCH T] ASSICNM COMPUTERS
\GNMENT
IDISCRETE CONTROL.) PANEL SYSTEMS
SYSTEM
} TO SECOND
ANALOG MODE | INTERFACE
CONTROL >
[ANALOG FUNCTIONAL .
CONTROL
Chart No. &
Now the interface part of the hybrid system is a little more complex than you might at first think.
When you say interface, most people say, oh well that is just some A-D converters, and some D-A
converters. Most certainly the A-D converters and D-A converters are a basic necessity to any inter-
face. But in addition to that, we feel you must have some general purpose logic as part of the inter~
face system. Now, this is necessary to control the signal flow between the digital and analog computers.
It is also necessary to supplement the amount of logic which is presented on the analog computers. In
the problems that we are considering for this particular system, that logic on the analog computers is
simply not adequate, It also contains a discrete control system which controls the signals going back
and forth between the computers. Actually it enables one computer to communicate and control the other.
There is also the analog mode control. Now this enables the digital computer to control the mode of the
analog computer -- that means hold, pot set, integrate, and so on, This is also the analog functional
control. This enables the digital computer to do such things as set potentiometers, readout amplifiers,
and so on. In order to make proper use of all this equipment, we think that there should be a patchpanel
on the interface. Now this enables us to make proper use of the logic and to correctly and efficiently
monitor and direct the discrete control system. The other function of the interface is the distribution
and assignment of the signal flow between the analog computers and the digital computers. We want to
have some flexibility and not have one analog computer always tied to a single digital computer. There
may be reasons for putting three analog computers with -one digital computer or even all four of them
with one digital computer.
ANALOG TO DIGITAL CONVERSION SYSTEM
TRANS- | ERROR |RESOLU-| S/H S/H LOGIC |CHANNELS| MUX | ASSIGN-| RATING
MISSION | (%R) | TION DECAY |RECOVERY! CONTROL WITH S/H |CHANNELS) MENT
i be (NO. BITS) (u/s SEC) (4 SEC) (TOTAL) | (TOTAL) |CAPABILITY
mM
ASTRODATA lo2 | .0232| I5 2 40 G 32 128 G |
(COMCOR)
BECKMAN 488 | .025 5 10 80 G 40 120 E 3
EAI (I) 309 | .04625| I4 33 40 P, 64 128 Po 2
EAI (2) 309 | .04625, I4 33 40 P, 32 64 P3 4
E EXCELLENT
G GOOD
S SATISFACTORY
P; POOR
~ P, S/H AMPLIFIERS CAN NOT BE DIRECTLY
CONTROLLED FROM INTERFACE OR ANALOGS
Pp “uv” BOX
P3 HARDWIRED
Chart No. 9
Now let's examine some of the technical features of the interface. In particular, the analog-digital
conversion system. We are interested in such things as transmission range, this is the time in micro- |
seconds it takes to transmit the data on 16 channels from analog to digital. The times given the Astro-
data system are quite a bit faster than the other two. The error as given here is a percent of reference.
Both Astrodata and Comcor are quite good, maybe twice as good as EAT. The resolution, that is the number
of bits in a converter, two of them at 15, two at 14. Sample and hold decay is measured in microvolts
per microsecond, and we would like to have a small number here so that it decays slowly; Astrodata at 2,
Beckman at 10, and EAI at 33. The sample and hold recovery time in microseconds -- we want this to be a
small number, We want it to recover rapidly. Astrodata and EAT at 40 and Beckman at 60. Logic control --
we think that both Astrodata and Beckman are quite good. EAI is rather poor. This is primarily because
in the EAI system, the sample and hold amplifiers cannot be directly controlled from the interface or
from the analog computers, but only from the digital computer. This puts some limitation on our programming
ability. Now we have the total number of channels which have sample and hold amplifiers. We are talking
about the total number on both systems. We have 32, 40 and 64. We requested 32. The Beckman units come
in groups of 10, s0 they took the next largest module that they make. EAI offered us 64 because they
claim that for them it was cheaper to offer us double the number rather than build flexibility into their
two interfaces. What we requested was that unused sample and hold channels from one interface could be
borrowed and used by the other interface. They did not have this flexibility. Instead, to make up that
requirement, they said ‘we will offer you twice as many . In their second system, they offered us 32.
Incidentally, the reason we had the two EAI systems here is that they actually sent in two bids. The
first bid was very responsive to our Request for Proposal, but they felt that the overall price was kind
of high, so they sent in an unsolicited second bid which is closer to their standard system. It doesn't
have some of the things we want, but we certainly felt we should consider it. The next channel shows the
total capacity of the multiplex system. Now, not all of these channels are currently available, but this
is the size of the multiplexer available for ultimate expansion and you see that the three systems are
roughly comparable, with some limitations on BAI's second system.
Assignment Capability- Now this is the ability to route analog signals to the two digital computers.
This gives us flexibility as to which computer is used with which analog machine. The Beckman system was
excellent in this regard. The Astrodata system was very good. EAI was poor. Thia is because in the system
they have here, the only way you can do this is by plugging and unplugging cables on what they call a
configurator panel. The other two do this by switches and electronic circuitry. The other EAI system has
no flexibility at all. Certain digital computers are permanently wired to certain analog computers. It
seems that EAI, when they conceived of the hybrid system, conceived of one digital computer working with
one or more analog computers, but they never thought that anyone would want a system involving two digital
Chart No. 9 (con'td)
computers, hence the lack of flexibility in this area. When we look at the overall picture and make
a@ comparison rating of the systems offered by the various competitors, we find Astrodata on top, with
the first EAI system second, the Beckman system third, and the other EAI system fourth.
DIGITAL TO ANALOG CONVERSION SYSTEM
ASTRODATA/COMCOR
BECKMAN /SDS
EAI (1)
EAI (2)
INTERFACE
OS > SOD eh % ~
Ss eS / SL S/ze 2X
Bgol € /88/ 88 Bio og / 98/82/85) &
Ses) § /S8/ 8 /SEY/ SE/ FF /FE/EE/ F
REY §& /P&/XS/REY ~F/ -S/ &X/ PEL _*
370).014| 15 | G | 5o| s | 6 | 6] 6G] t
508 |.02 | 15 | 6 | 300] S$ 60 | — | 3
367 |.025| 14 | s | 50] S | Bm | 64 | Ps | 2
367 |.025 | 14 | Ss | 50| AR} BR] 4 | Rh | 4
E EXCELLENT
G GOOD
S SATISFACTORY
P. POOR
P, NO INDIVIDUAL OVERRIDE CONTROL
P, DATA CAN NOT BE LOADED WITHOUT THE
USE OF THE DIGITAL COMPUTER
P; ‘J’ BOX
P, HARDWIRED
10
Chart Ne. 10
Now we look at the digital to analog conversion system. Again we are interested in transmission rate.
Now this is the time taking to tranemit the data on all 32 channels, and it is measured in microseconds.
Beckman is slow compared to the other vendors. Error as a percentage of reference ~- this is the overall
through put error. Astrodata is a little better than the others. Resolution, total number of bits --
again Astrodata and Beckman working with 15 bit units - EAI with a 14 bit unit.
Analog Compatibility «- We rate Astrodeta and Beckman both good, and EAI satisfactory. This is because
both of these companies use computing amplifiers for the output amplifier of the digital to analog
converter. EAI uses an amplifier supplied by Adage. It has different characteristice. It doesn't have
as much current output. It doesn't have as low an output impedance, and various other characteristics.
Analog recovery time is in microseconds. Beckman is quite a bit higher than the others.
logic Control - They are all satisfactory, except the second EAI system. Now this is because that system
does not allow any individual override control of the digital computers -- which puts some limitations on
our programming ability.
Local Control - Astrodata and Beckman are good. EAI is poor. This is because in the EAI system, data
cannot be Loaded into the DAC's without the use of the digital computer. Now when the problem is running
normally, this is no limitation, but during the trouble shooting phase and for maintenance operations, this
would be a serious handicap.
Total number of channels - Everyone is in the 60's.
Assignment Capability - This is the same sort of thing we talked about on the other chart. Again the
same types of ratings for the same reasons. When we rate the vendors on their digital to analog converting
systems, we find that Astrodata is in first place, EAI second, Beckman third and EAI fourth. It's the
same as the previous chart.
GENERAL PURPOSE LOGIC
ASTRODATA/COMCOR
BECKMAN (SDS)
EAL (1)
EAI (2)
TYPE BASIC LOGIC
(SYNC. LOGIC CLOCK CONTROL
QUANTITY OR FREQ. FREQ. AND
(LOGIC UNITS) ASYNC) (MP/S) (MPS) | MONITORING] RATING
454 BOTH | 0.1 S 2
(1 MP/S)
526 ASYNC 5 0.1 S 3
5 \4 SYNC 20 O. | E 1
(2MPIS)
O — —- — — 4
E- EXCELLENT
G- GOOD
S - SATISFACTORY
i
Chart No. 11
Now I want to talk to you about the general purpose logic which is on the interface. The quantities
are all adequate for our purposes, The type of logic could be either synchronous or asynchronous.
Astrodata offers us the option of having both synchronous and asynchronous logic. That is, we can make
a portion of it synchronous, and we can use the rest asynchronously. Beckman offers us only asynchronous
logic. EAI offers us only synchronous logic.
Basic Logic Frequency - This refers to the ability of the individual logic elements to respond to a
signal. Astrodata works at 1 megapulse per second; Beckman at 1/2 megapulse per second, EAI at 20 mega-
pulses per second. The clock frequency, that is the frequency at which the logic is clocked, in all
cases, is 100,000 pulses per second.
Logic Control and Monitoring - EAI is a much better system than that offered by either Astrodatea or
Beckman. in the second EAI system, they just didn't offer any logic. The comparative rating on this
particular element of the interface, EAI 1s the winner, Astrodate second, and Beckman third.
DISCRETE CONTROL SYSTEM
@ ASTRODATA/COMCOR |__ 136 136 | 48 G
@ BECKMAN /SDS_|__112 112 | 46 G 2
@ EAI (i) 128 128 80 p* 3
@ EAI (2) O O 0 - 4
* NO MASK CAPABILITY PROVIDED
Ie
Chart No. 12
For the discrete control system, we have sense lines, control lines, and interrupt lines, and we have
adequate quantities of each. Now the sense lines are used to keep the digital computer informed of the
state of the analog problem. The control lines are used to control analog computer functions. The
interrupt lines are used to send interrupt signals from the analog computer to the digital computer,
and the interrupt control is how we control the interrupt lines. We have to grade EAI poorly here,
because they had no masking capability. In the comparative ratings of this particular section, we
have to give - Astrodata, Beckman and EAI.
INTERFACE RATING SUMMARY
ANALOG TO |DIGITAL TO | GENERAL | DISCRETE
DIGITAL CONV.\ANALOG CONV) PURPOSE | CONTROL | OVERALL
SYSTEM | SYSTEM LOGIC SYSTEM | RATING
@ASTRODATA/COMCOR| 1 { 2 |
@ BECKMAN /SDS 3 3 3 2 3
@EAI (1) 2 2 | 3 2
@EAI (2) 4 4 4 A A
A
Chart No. 13
In the interface rating summary, we take the quantities we've looked at before and give you a summary
chart. In the overall rating for the interface, that's considering the A-D converters, the D-A converters,
the general purpose logic, and the discrete control system, in summation we think Astrodata has the best
one, EAI the second best, Beckman third, and the second EAI system runs a poor fourth.
DUAL CDC 3300 CONFIGURATION
16K
CORE
24 BIT
16K
CORE
16K
CORE
INTERFACE
3304
CPU
TAPE CONTROL- LINE
UNIT DISPLAY PRINTER
60 KC CONSOLE 1000/MIN
DISK CARD CARD
3.3MEG READER PUNCH
CHAR 1200/MIN 250/MIN
TAPE CONTROL:
UNIT DISPLAY
30KC CONSOLE
3304
CPU
INTERFACE
[4
Chart No. 14
Now I will talk about digital computers. The first of these will be the configuration offered for the
CDC 3300. This has two central precessing units, three banks of core, each bank having 16,000 words, and
each word being 24 bits. Each uf the CPU's can talk to any or all of the banks of core. They share a
disk with 33 million characters and they share the normal digital peripheral units, the card reader and
card punch. There are two tape units, one to each CPU, two control display consoles, one to each CPU and
the line printer is constrained to operate with only one of the CPU's. Each of the CPU's is connected to
an interface which in turn is connected to the four analog computers.
SDS 9300 CONFIGURATION
loK
CORE
24 BIT
CONSOLE
16K
CORE
OPERANDS
ONLY
loK
CORE
INTERFACE
CONSOLE
9300
CPU |
SETUP CARD TAPE
CONSOLE] [READER UNIT
|
rarer PROG.
CHAR. SW. I 1
: LINE CARD CARD
PRINTER| |PUNCK] — READER
JO00/MIN) -[300/MIN}-——_JAOO/MIN
I
SETUP CARD TAPE
9300 CONSOLE] READER UNIT
l i
CPU
INTERFACE
Chart No. 15
On the SDS system 9300 digital computer, again we have two CPU's connected to three banks of core storage.
Each of the banks is 16,000 words, they are 24 bit words. However, you will note that this CPU does not
have complete ability to talk to both of these banks of core. This CPU has complete freedom to this core,
but can use this cere for operands only. In no case can either of the CPU's use all three banks of core.
It shares a disk of 16 million characters and it shares the peripheral's through 4 programming switch.
Now this enables all of the peripherals to be used by either one computer or the other. You have to
actually throw the switch in order to switch them back and forth, and they 811 switch at the same time.
In addition, we have the operational consoles, and a one tape unit to each CPU. Additionally we have
set up console and card reader. Now these are smaller items and are made by the Beckman Company for use
with the 9300 when its part of their hybrid system. These are specialized read out and control units for
hybrid work and there is one set with each CPU. Both of the CPU's are attached to interfaces which are
again in turn attached to the analog computers.
EAI 8400 CONFIGURATION
=—— 2wo BID DELETION
loK =——_ Zuo BID ADDITION
CORE INTERFACE
32 BIT
CPU
loK | | CONTROL
CORE CONSOLE éc/0
CONSOLE
DISK S HTAPE 3 THE
3H a HTaPe CARD CARD
I5.3mMEG] [SE we READER PUNCH PRINTER
CHAR. F © MTAPE! ES Tape 800/MIN 100/MIN 007M
|
l6K CONTROL
CORE L_] CONSOLE ¢ C/O
CONSOLE
CPU
l6K
CORE INTERFACE
Chart No. 16
Now for the EAI 8400 Digital Computer. Again we have two central processing units. In their first bid,
they offered to tie each of the CPU's to twe banks of core memory, each bank of memory having 16 thousand
words, and the words here have 32 bits. In the second bid, they removed one bank of core, but they did
connect the three cores together so that each CPU could talk to all three cores. The two CPU's share a
15 million character disk, and they share the normal digital peripherals, a card reader, a card punch, a
line printer, either a 1000 line a minute, or a 600 line a minute printer between the two CPU's.
The original bid had 4 tape units, the second bid removed two of the tape units. We also have the
control consoles and the special hybrid control consoles. Each of the units are connected in turn to
interfaces which are then connected to analog computers.
CDC 6400 CONFIGURATION
PP |
DISK pp pp
BA MES. sn or MONITOR | INTERFACE
pp pp
> TAPE
UNIT PN 6405 CPU a
60KC é
CENTRAL MEMORY
32Kx 60 BIT
CARD PP. pe
READER INTERFACE
1200/MIN.
PP 10
CARD PP CONSOLE PP
PUNCH
250/MIN.
PRINTER CRT CONTROL
Chart No. 17
Now in the course of evaluating digital computers for the hybrid system, we came across the CDC 6400
computer. This is a change from our original conception of using two central processors. This has a
single central processing unit and a single central memory. However, this memory is 32 thousand words
and they are 60 bit words. We also have ten peripheral processors which work with the central memory.
Each of these peripheral processors has an additional 4 thousand words of memory and they are 12 bit
words, We have a set of peripherals and an 44 million character disk, 2 tape units, a card reader,
punch and line printer, and we have the cathode ray tube displays and controls. These are remote input /
output, on-line control units. We have the main control console. Now, in practice, we would assign four
of the peripheral processors to work with the various digital peripheral units, and we would assign another
four peripheral processors to work with the two interfaces. One processor would be assigned strictly to
the main control console and the other processor would contain the monitor for the overall operation of
the system.
The digital computer then is connected to the two interface units which are then connected to the analog
computers. Even though we consider a single digital computer, we still kept the concept of two separate
interfaces simply because of the size of the units, the amount of equipment in them, and the size of the
patchboard.
DIGITAL COMPUTER
STORAGE PERIPHERALS
pe CORE DISC CARD
SPEED SIZE MILLIONS | TAPE. | READER | PRINTER | PUNCH
MICRO BITS/K STOR. | CHAR. CPM LPM | CPM
CDC 6400 | 1.0 60 x 32K+ 84.1 2 | 1200 | 1000 | 250
12 x40K
EAI 8400 | 2.0** 32 x48K (5.3 2 800 600 100
EAI 8400 | 2.0** 32 x 65K 15.3 4 800 1000 | 100
Ist BID |
CDC 3300 | 1.3** 24 x 48 33.0 2 1200 | 1000 | 250
SDS 9300| 1.75 24 x 48 * 16.0 400 | 1000 | 300
* NOT ADDRESSABLE FROM BOTH CPU'S
**
SLOWER WHEN STORAGE REFERENCE UTILIZES
RELOCATION HARDWARE
18
Chart No. 18
Now let's look at some of the performance characteristics of the various digital computers. Let us first
look at speed memory cycle time, The speed in microseconds for the 6400 is 1; EAI is 2; and 3300 is 1.3;
and SDS 1.75. Now, these units are somewhat slower when storage reference utilizes relocation hardware.
This is what is known as a paging system, and it is important when you have more than one central processing
unit working with a common memory. You want to protect your memory so that one computer dees not write
in the other computer's memory. On the CDC 6400, I think it is worth pointing out that the effective
eycle time can be reduced to .1 microseconds. This is due to memory overlap techniques which are possible
in thés machine.
Core size - The 6400 has 60 bit words and there is 32 thousand of them. The peripheral processors have
[2 bit words, each of the ten have 4 thousand words for a totel of 40 thousand. EAI uses 32 bit words
and they have either 65 thousand words or 48 thousand, depending on how many banks of storage we consider.
The 3200 has 24 bit words with 48 thousand words and the same for the 9300. However, in the 9300, we mst
remember that all of the storage is not addressable from both of the central processing units. The disk
units associated with the various devices in millions of characters, 64, 15 and 33 and 16, Now we feel
that the minimum disk size that we can use is in the neighborhood of 30 million characters, which means
we have to either compromise our programs to use these disks, or have additional units quoted by the vendor.
This one (6400) seems to be somewhat in excess of our minimum requirements, however, it is the smallest
disk that ig offered with that particular computer system. We figure two tape units should be adequate
for the system that we are contemplating. Standard digital peripherals -~ you get what you pay for here.
DIGITAL COMPUTER
CDC 6400
EAI 8400
EAI 8400
CDC 3300
SDS 9300
FLOATING WG |
ACCUR| |. EXEC. DATA TIME SHARING | INTERRUPT
DEC DIG} SPEED |REL6400| CHANNELS | CAPABILITY CAPABILITY
x
14 S+ 1.0 12 S+ Sr
7 S+ 2.55 4 S+ S+
7 S+ 255 4 S+ S+
10 S 3.08 8 S+ P
11 S 4.25 8 P S+
* MACHINE CONTROLS CPU FROM PP RATHER THAN INTERRUPTS
19
Chart No. 19
More characteristics of the digital computer. In floating point, the accuracy expressed in decimal
digits. We have 14 decimal digits on the 6400, 7 from the 8400, 10 from the 3300, and 11 from the 9300.
In floating point speed, they are all satisfactory. Some are somewhat faster than the others. If we
take the execution time of a Fortran compiled program, we find that if the Control Data 6400 can do that
in a certain length of time, which we designate as 1, it will take the &400 24 times to execute the same
program, and the 3300, three times as long and the 9300 more than 4 times as long.
fhe number of data channels available are 12, 4 and & Im time sharing capability of the computers, all
of them except the 9300 are adequate. In interrupt capability, the 6400 and the 8400 are very good in
this regard. The 3300 is not adequate. In the 6400, the peripheral processors act as a buffer between
the outside interrupt signals and the central processor.
POWER OF VARIOUS COMPUTERS
RELATIVE TO CDC 6400 vs INTERRUPT FREQUENCY
RP |
0.4
0.3
0.2
0. |
f=INTERRUPT FREQUENCY
IPT=INTERRUPT PROCESSING TIME
P=STATIC RELATIVE POWER
RP =P (I-f xIPT)
— EAI 8400
——-CDC 3300
=——SDS 9300
0 | 2. 3 4 5 6 7
f IN KILOCYCLES
2.0
Chart No. 20
Now when we compare the power of the various computers with respect to interrupt processing capability,
we come up with some interesting numbers. On this particular chart, we used the 6400 as a base, since it
is the fastest of the computers, and if we had the data from the 6400 displayed on this chart, it would
be way up here someplace. The number on the scale would be 1.0. The line would go out to the right with
& very small negative slope. It would actually intersect the axis out here at around 250 kilocycles.
You can see that the basic speed difference between the 8400 and the 6400 is about 24 to 1. The difference
with the 3300 is about 3-1 and a little under 5-1 for the 9300. When you get out to around 5 kilocycles,
which is what we think will be the frequency at which we are commonly processing interrupts, you will see
that all of these drop rather drastically, and they all bunch together in an area which indicates that
the 6400 is roughly 5 times as powerful as these machines in processing interrupts. Now, it must be
remembered that we will have two of these machines in the system, and only one 6400, but even with two of
them, it means that the 6400 is now 23 times as fast in processing interrupts as two each of these
computers.
SOFTWARE
S+
S+
S+
St
S+
S+
St
S+
S
S+
p
S+
6400 CDC
8400 EAI
3300 CDC
9300 SDS
21
Chart No. 21
A very important consideration in a digital computer system is software -- the availability of it and
quality of it -- and when we look at the basic monitor ~- we are talking about a real-time monitor here -~
we find that they are all satisfactory except the 3300, and we know that isn't satisfactory because we
have a 3200 in B/151 and we have had to write our own real-time monitor for it. The Fortran system ~
They are all satisfactory again except for the 3300. This is not as compatible to existing programs
which are now running in Fortran on the 7094 as are the others. We hope that for many cases the digital
portion of the hybrid problem which we contemplate doing, that we would be able to make use of at least
part of the existing programs which have been written for the 7094. In hybrid support software -- EAL
is clearly superior to all of the others. They have put a lot of time and money and effort into this.
Now, although none of this software is completely written, debugged, and available today, the program
has been underway for more than a year. A large number of people are working at it. We have talked to
these people. We have a very high regard for their ability and we're quite positive that the software
will be available before the computer is delivered. There is very good hybrid support software for the
9300. This is primarily due to the fact that the Beckman Company, in cooperation with SDS, has been in
the hybrid computer market for the past several years. Both of the CDC units are poor in this respect.
They do not have any background of hybrid software. In application software, again we are speaking of hybrid
application software, the same story holds. Time sharing software, the 6400 and the 8400 are both
satisfactory in this regard, particularly the 6400. It was designed for time sharing. The 3300 and 9300
are poor. Diagnostics routines are all satisfactory. Disk oriented software -- they are all alright
except the $400 which, when it was originally conceived, did not contain a disk, and consequently the
iisk software has not yet been written.
The selection order of the digital computer system based on software, particularly because of the emphasis
that we give the hybrid software, EAI is definitely the leader, closely followed by the Beckman-SDS
system. The CDC units are third and fourth.
MAJOR EXCEPTIONS
MAJOR EXCEPTIONS
WEAKNESSES
EAI 8400
FLOATING POINT ACCURACY ONLY
7 DECIMAL DIGITS
CHANNEL SWITCHING UNIT FOR THIS SYSTEM ONLY
MEMORY PROTECT SYSTEM FOR THIS SYSTEM ONLY
DISC CONTROLLER UNIT FORTHIS SYSTEM ONLY
DISPLAY CONTROLLERS UNIT FOR THIS SYSTEM ONLY
PERIPHERAL EQUIPMENT OF MANY MANUFACTURERS
ALL SOFTWARE PACKAGES UNDER DEVELOPMENT
CDC 3300
FORTRAN INCOMPATIBLE 7094
NO TIME SHARING SOFTWARE AT TIME OF DELIVERY
NO APPLICATIONS SUPPORT SOFTWARE
S0S 9300
PROCESSOR CORE ADDRESSING
PERIPHERAL ACCESS POOR
NO RELOCATION HARDWARE
NON STANDARD CONFIGURATION
CDC 6400
*ASTRODATA HAS UNDERWAY A COMPANY SPONSORED PROGRAM TO DEVELOP APPLICATIONS SUPPORT SOFTWARE FOR THE 6400. THIS ISA
15 MAN YEAR PROJECT
NO APPLICATIONS SUPPORT SOFTWARE*
ae
Chart No. 22
Now let us look at the summation of the major problems associated with each of the digital
computers. EAI's unit = Their big problem is floating point accuracy at only 7 decimal digits.
The 9300, their big problem is the fact that both CPU's cannot address all the core. There
is a limitation on how much core each CPU can address. Other problems, but not quite so
significant - in the EAI system = the channel switching unit, the memory protect system,
the disk controller unit, and the display controllers unit are offered for the Lockheed
system only. This is not their standard hardware. They proposed to do special engineering
and develop these units only for us. This causes us some uneasiness because we would much
prefer, if at all possible, to buy standard off-the-shelf hardware, that they are going to keep and
sell to other people. Then we know that it will be supported in future years and some of the
uncertainties are taken out of it. Another thing, the digital peripheral equipment for the
EAI computer is obtained from many different manufacturers, Now EAI has entered the digital
computer field rather recently, and they make their own main frame and they have designed the
system.
They have picked up peripheral equipment from all and sundry. Their printer is made by
one company; thelr tape reader by another, the card reading and punch equipment by still
another. We feel that some of the larger digital computer companies whe make most or all
of their own peripherals are a little bit more consistent. All the software packages from
EAT are still under development. They're a new company in the field and they just haven't
had the time to develop software that CDC and SDS have. In the CDC 3300, we have Fortran's
incompatibility with the 7091. We have no time sharing software available at the time of
delivery. We have no hybrid applications support software. In the 9300 we have, relative
to the others, relatively poor access to the digital peripheral equipment. We have no
relocation hardware. That one core that both of the CPU's can get at, we will have to
take extra pains in programming, It's a non-standard configuration and that requires
special engineering and problems may crop up of which we are not yet aware. In the 6,00,
the major weakness is no application support software. However, this is a midigating
circumstance here. Astrodata, whic