REPORT OF TIME-SHARING SYSTEM SELECTION COMMITTEE
Document Type:
Collection:
Document Number (FOIA) /ESDN (CREST):
CIA-RDP78-03948A000100090001-8
Release Decision:
RIPPUB
Original Classification:
S
Document Page Count:
66
Document Creation Date:
December 16, 2016
Document Release Date:
May 25, 2005
Sequence Number:
1
Case Number:
Publication Date:
June 26, 1968
Content Type:
MF
File:
Attachment | Size |
---|---|
CIA-RDP78-03948A000100090001-8.pdf | 2.65 MB |
Body:
App ove For Releaae /06/ CIA-RDP78-03948A000100090001-8
26 June 1968
MEMORANDUM FOR: Director of Computer Services
SUBJECT : Report of Time-Sharing System Selection
Committee
25X1A
2. The applications' division chiefs assumed responsibility for
ascertaining projected time sharing workload. The results of these
surveys are attached to this report as Attachments A, B, and C.
25X1 A
4. Compatibility of the selected system with the present OCS sys-
tems was stated as an essential objective by No 25X1A
other members of the Committee disagreed with this objective. In effect,
this objective narrowed the selection to an IBM 360, or possibly the RCA
SPECTRA 70/46. Attachment E contains a list of other systems reviewed
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
_9CIIT
Approved For Release =t~VX%gk7DP78-03948AO00100090001-8
25X1A
and the primary reason for rejection from further consideration. The
systems studied in depth are:
a. TSMON - 50, the current OCS time sharing system.
b. TSMON-RUSH - 50, the current OCS time sharing system
with LCS (large core storage).
c. ADEPT - 50, the SDC (System Development Corporation)
system which is under consideration by IPRD.
d. TSMON - 65, current OCS system on a Mod 65.
e. TSMON-RUSH - 65, current OCS system with LCS on Mod 65.
f. ADEPT - 65, the SDC system on a Mod 65.
g. TSS - 67, IBM's time sharing system for the Mod 67.
h. CP - 67, a virtual machine system written by Lincoln Labs
and IBM for Mod 67.
i. TS - 70/46, a system by RCA written for the Spectra 70/46.
j. Duplexed small machines such as twin Mod 40's or twin
S-70/46's.
See Attachments F, G, and H for cost comparisons.
5. Widely divergent views are held by the six Committee members.
A general characterization of the views is that the three applications
oriented members have conservative views and wish to delay time sharing
and put more emphasis on improvement of the batch system OCS is cur-
rently using. However, specific reasons, objectives, and emphasis
appear to vary considerably among the above three members.
believe in time sharing and believe that the Agency
should progress rapidly to a fully implemented system; however, these
three members also have differing opinions as to specifics. These
three have come closer together in views, since the favorable perfor-
mance of the CP-67 on comparison performance tests and most of the
specific differences are trivial as to long range implications. -
2
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
25X1A
mr) TN
(eq
Approved For Releas
IA-RDP78-03948A000100090001-8
F
25X1A
25X1A
6. Since has promised the Director of Computer Services
a completion of the study and a recommendation as to future direction
by 1 July 1968, and since there is little probability that a consensus of
opinion will emerge, the following course of action is taken.
a. This paper is being written and includes a general discus-
sion of the selection considerations. It also includes statistics on
performance comparisons. 0 has other statistics and infor-
mation available on systems.
b. This paper does include a recommendation by
Committee Chairman.
0
c. This paper is being presented to the Committee on 26 June
1968 for review.
d. Each member is then being asked to write for the Director
of Computer Services a concurrence in the recommendation or a
dissenting memorandum detailing his objections and recommending
an alternate selection or course of action.
7. Objectives. The objectives of the study are those of the memo
"Planning for IBM 360/50 Replacement" to CSA by DD/OCS of 26 April
1968. See Attachment I. In addition to these objectives, security com-
partmentation especially between files accessable to Agency users only
and those accessable to non-Agency users is an overriding consideration.
8. Discussion of 10 Studied Alternatives.
a. IBM 360/65. Since the Mod 67 has hardware features such
as dynamic relocation and special channel functions which are
extremely desirable for time sharing and since the Mod 67 leases
for less than $5, 000 per month more than an equivalent Mod 65,
and since the Mod 65 cannot use such time sharing software as CP,
CMS, and TSS; and since the Mod 67 can run in a Mod 65 mode as
Mod 65 backup, all systems which included a Mod 65 were rejected.
This decision appears obvious, especially when considering the
extra very skilled manpower costs required to compensate for Mod
65 hardware time sharing deficiencies.
b. IBM 360/50 - RUSH. This system requires two million
bytes of LCS, will handle 90 users, but uses PL/l only and has no
query language. This system is good, but is very inflexible to
adopt to known and speculative future Agency needs.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
25X1A
eJ E 1.1
I
Approved For ReleaseBZ7DP78-03948AO00100090001-8
c. IBM 360/50 - ADEPT. Not operational yet, will support
nine terminals on Mod 50 and 30 on Mod 65. LISP, JOVIAL, TINT
(JOSS type language), TDMS (CAPRI type file management system).
Requires drum - does not appear to have an effective scheduling
algorithm. This system possibly could be used in an experimental
environment but it was felt that OCS has advanced beyond this level.
Production must be mixed with experimentation.
d. IBM 360/50 - TSMON. Several possibilities are available.
LCS (Large Core Storage) could be added. However, partition sizes
are limited and this relegates the customer to use languages developed
by the systems programmers. Procedural languages would use
excessive amounts of core in this environment. Although OCS senti-
ment and pride tends to bias a decision towards this system, it was
felt that much more flexibility is required to meet the future unknown
requirements. Also, the good features of this system could be
salvaged for another system with a minimal amount of reprogramming.
Maintenance of and feedback to this system depends solely on Agency
effort; no other users are contributing to the research and further
development.
e. IBM 3.60/67 - TSS. Not fully operational and performance
is very poor.
f. TS - RCA_ 70/46. This system has many fine features. RCA.
has learned from the mistakes of others. RCA answers every objec-
tion with a promise of support or performance. However, RCA
promises too much and shows too little which is working. Also, a
technical evaluation of RCA's promises of performance and support
suggests that RCA is promising the impossible, especially for the
January 1969 target date. RCA. promises direct compatibility or
their effort to make files, programs, and hardware compatible to
present systems. If these are true, it will be easy to convert to
RCA (which is much, much less expensive) at a later date. The
progress and performance of the RCA 70/46 should be reviewed and
compared after it is running. Thus, for the present, the RCA 70/46
has been rejected. (See Attachments G and J for additional reports
on RCA 70/46.)
g. Duplexed Small Systems. RCA 70/46 and Mod 40's were
considered. The big advantage in these cases is binary security
compartmentation. This is absolute and is approved by the Office
of Security. A secondary reason is backup, however, this duplex
backup would be at the expense of having both security compartments
active. If the RCA 70/46 were tested in the future and approved
Approved For Release 2005/Tte"
JEQ
11
Approved ForZelease 2005/06/03 : CIA-RDP78-03948 000100090001-8
technically as an adequate time sharing system and if the price
remained as it is at present, it should be considered strongly in
the future. The Mod 40's are not powerful enough to allow for any
expansion or progressive time sharing development. To box the
Agency time sharing efforts with such constraints is extremely
shortsighted. If the Office of Security insists on separate hard-
ware systems, consideration should also be given to a Model 40
for the COINS network and to retaining TSMON on a Model 50 for
internal Agency time sharing needs. (See Attachment K for a
proposed configuration.)
h. CP-67 (IBM 360/67). This system is working, is avail-
able, and meets all objectives of the study as outlined in the DD/OCS
memo (see attachment I). Negative attributes are cost which is high
(but within the planned budgetary limits), and the objections of the
Office of Security to putting all material on one physical piece of
hardware. CP-67 is running operationally at Lincoln Labs, Wash-
ington State University, and several others. All are converting
rapidly to sole use of CP on the Mod 67 for their time sharing sup-
port. Most of these environments are not similar to that in OCS;
however, Lincoln Labs has a similar enough load and mix to draw
comparisons and conclusions. LL has 30 terminals on-line with
several virtual machines in the background using OS and CMS.
Terminal response time is excellent and the CP will run a number
of operating systems such as CMS, OS, DOS, TSMON, ADEPT, etc.
and concurrently if needed. CMS has a text editor, FORTRAN-G,
ALC, PL/l, and other languages on-line. The OCS language SOLVE
is a sub-set of PL/l, and SOLVE problems were run correctly under
CMS by the simple addition of a semicolon at the end of each state-
ment. This PL-1 is not an incremental compiler and thus must be
compiled as an entity. For some purposes, SOLVE has advantages
over the PL-1. CP-67 was previously CP-40 and its experience
level dates back more than two years.
The greatest advantages are its flexibility and power for expan-
sion. Production and experimentation can be processed concurrently.
See attachments D and H for additional details.
A comparison of CP-67 characteristics with objectives of the DD/OCS
memorandum follows:
a. CP-67 and its CMS is a good time sharing system with fast
response. It has an excellent background processing capability
which is 100% compatible with other OCS processing.
b. CP-67 comes very close to meeting a general objective of
both multiprogramming and time sharing. This is not recommended
Approved For Release 2005/06/03 :5C R-RD~P78d-0772$Aj$~0100090001-8
Approved For Release 2005/06/0a.: 1A ft
at this time, but the feasibility of doing this operationally should
be investigated. Multiprocessing is not available in the strict sense
as of this date. CP-67 is compatible with present systems in that
other systems can run under it.
c. CP-67 not only can provide a production time sharing en-
vironment by January 1969, but also an experimental one by making
each a virtual system under CP-67.
d. Experimentation and extensions can be done on the same
system by the virtual machine extension.
e. Costs are within the budget. Even should additional costs
arise (unlikely unless to take more advantage of background capa-
bility by having more I/O devices), this system is so superior to
others that justification should not be a problem. As a temporary
economic saving the drum (2301) could be omitted and a performance
measurement taken without it. This is only a second rate alternative,
however.
f. Conversion is fairly easy. TSMON can run under CP as soon
as several new peripheral support modules are finished (momentary).
This probably is not desirable for the long haul, but the system
flexibility does not create any extreme time constraints or pressures
to convert.
g. Load - CP-67 has the power to expand greatly. Of course,
every expansion will degrade background capability.
h. Manpower Resources. Many time sharing systems have been
written for IBM 360. All of these systems can be tested as virtual
systems under CP, and the developed packages can be extracted or
run unchanged under CP-67. No other system gives us as much
software flexibility, expansion as CP, and thus CP allows more
human resources to be placed on solving customer problems. It is
strongly stressed that CP is a large and powerful system. New devices
and new functions will be desirable and will be added. OCS must plan
to staff this effort with an adequate number of system programmers,
9. Security Considerations. Two separate machines for non-Agency
users and files and others would give the Office of Security comfort. CP-67
does not meet this requirement in that it is one physical machine.
However, CP-67 has a new concept, "virtual machines" and it has
special hardware to protect virtual machines from each other. The pro-
tection given in the virtual concept probably has a higher probability of no
Approved For Release 2005/06/0 WET ZRR 03948A000100090001-8
Approved For Release 200
~~~~~ =-78-03948AO001 f0'600090001-8
error than electronic communication systems which are used in Agency
message handling and which the Office of Security allows to function.
Expediency, i. e. , data communications, has overcome Office of Security
reluctance to approve. The time is approaching when a review of com-
puters and their compartmentation security is in order. Just as in com-
munications, the time will be reached when maintenance of separate sys-
tems will not be feasible.
However, let us assume that the Office of Security will not approve,
at least at this time. It is recommended that the system service non-
Agency personnel for a block two hours a day. Agency users who were
not using privy Agency files could still function on-line, such as SOLVE
users, program editors and conversational programmers. During the
remaining 22 hours a day, Agency users could have free access, and
this access could include those files which were shared with non-Agency
users during the two hour block time.
As an alternative recommendation, the Mod 40 in CRS could service
non-Agency users, i. e. , be the COINS processor. Generally, the files
on this system are those which are available to the non-Agency users of
COINS. Those CRS processed files which are specially sensitive could
be processed by OCS. It would seem that the administrative, manage-
ment, and technical details would not provide unsurmountable obstacles.
As an alternative to replace either of the above at a future date,
research should be undertaken to test ROS (Read Only Storage) as an ab-
solute protection for one of the virtual machines. This seems feasible,
especially on the Mod 67, however, it would take time to develop. Pos-
sibly, mid 1969 or early 1970 can be suggested as target dates for this
evaluation.
10. Conclusion. It is recommended that the Agency lease a Mod 67
of the configuration shown in Attachment H for January 1969 to replace the
present Mod 50. It is further recommended that Advanced Projects Staff
put together a conversion plan for this changeover. Acquisition of such a
system is a positive and progressive approach to solving Agency computa-
tion and information handling problems. Intuitively, such a step will
prove to be the most economical over the future years.
It is also recommended that the letter of intent to IBM for lease of
the IBM 360/67 clearly state that support for CP and CMS software must
Approved For Release 2
Approved For Release A5/067097DP78-03948A000100090001-8
be supplied. CP/CMS software is IBM Type-3 (same type as HASP) and
sometimes IBM will default on software support for Type-3 programs
unless this is requested before contract negotiation.
In summary the CP-67 (CMS) is recommended because:
a. It is seasoned; it has been working for two years.
b. It does handle many terminals with an excellent response
time. One installation occasionally has 60-80 terminals active,
and another has 20-30 active in a routine production mode.
c. It is 100% compatible with the OCS batch processing systems.
d. Its background processing capability is at least 50% of a
Mod 65, and depending upon human and procedural variables, may
be as high as 100% of Mod 65.
e. It does multiprogram.
f. It can be used as a production time sharing machine.
g. It can be used experimentally and concurrently with produc-
tion to test subordinate time sharing systems.
h. Costs are within the OCS time sharing budget.
i. It has the best internal compartmentation hardware of any
one machine, and has possibilities of being engineered to absolute
compartmentation.
j. Conversion problems are minor and less than any alternative
other than direct upward extension of the present system.
k. Expansion - It has the brute power to handle peak terminal
loads and quadrupling of projected load can be absorbed with no
serious degradation.
1. Upward compatibility - The IBM 360/67 is upward compatible,
either to duplexed multiprocessing hardware or to the IBM 360/85
(a version of this machine is being equipped with time sharing hard-
ware).
m. More software, both time sharing and batch types, is avail-
able and will run on this machine more than for any other machine.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
7.1
vised)
II LIT
Approved For Release 05/0 /03 : CIA-RDP78-03948A000100090001-8
n. Manpower requirements are less than for any other com-
parable system.
o. The better features developed for the OCS time sharing sys-
tem can be salvaged.
The five other members of the Committee will be giving D/OCS
memoranda of concurrence, comments, modifications and/or rejections
of these recommendations.
25X1A
Approved For Release 2005/06/039 CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
MEMORANDUM FOR: DD/COS
SUBJECT: Proposed SAD TSS Requirements
1. As of 1 July 1969
Application: Project Profiles
Number & Type Terminals: one 2260
DAS requirements: two million bytes
Program requirements - Core: Possibly one 15K module
for one special purpose program.
All other requirements can be
satisfied with LINUS, TSAR, etc.
2. As of 1 July 1970
A. Application: Project Profiles
Number & Type Terminals: one 2260
DAS requirements: two million bytes
Program requirements -
Core; Possibly one 15K module
for one special purpose program.
All other requirements can be
satisfied with LINUS, TSAR, etc.
B. Application: Project MEDSTAT
Number & Type Terminals: Two 2260's
DAS requirements: Two million bytes
Program requirements -
Core: No special purpose requirements.
Needs TSAR, LINUS, SOLVE.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
11
Approved For Release 2005/06/03: CIA-RDP78-03948AW696641-8 All
Page 2
C. Application:
,Cartography, ANDI applica-
tions, other SAD Ad Hoc require-
ments.
Number & Type Terminals: One 2260, one 2250.
Das Requirements: Unresolved at this point in
time. At least one million bytes
for Cartography by 1 July 1970.
Program requirements: 100K for CAM. 200K for SADIG.
Other requirements can generally
be satisfied via existing TS
programs such as LINUS, SOLVE,
DESKCAL, etc.
3. The proposals listed herein assume an OCS decision to
support time sharing. They do not necessarily 'require a time
sharing environment and with the possible exception o the
project profiles application might very well function better
in an environment in which the expenditures were placed on
decreasing turnaround time as opposed to buying and supporting
a time shared system. There is a need and the justification
in all cases for an RJE capability combined with the flexibility
to alter source coding and data via a remote terminal device.
But it is possible to incorporate these features in an MVT
environment therefore pzoviding at least a 25% decrease in
turnaround time for the whole of OCS assuming a one for one
dollar expenditure on equipment compatible with existing OCS
hardware versus the expenditure on time sharing, while at the
same time providing the flexibility on all OCS hardware of
the most valuable exting time sharing systems capabilities.
In summary, the real need is for:
1. a capability to run very large programs,
2. a remote job entry capability,
3. the flexibility to alter programs and data
through an on-line remote console, and
4. the ability to have output either printed
in hard copy or displayed on a 2250 or 2260 display
device and in the special case of the ANDI system
to have pseudo real-time output as opposed to a HASP-like
queued output system.
25X1
Approved For Release 2005/06/03 : CIA-RDP78-03948AOOQ100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Attachment "B"
PROJECTED TS APPLICATIONS from MSD
Estimate, TS Applications as of 1 July 1969
APPLICATION # TERMINALS' TYP'E' DA'SD
TYPE
CORE
MAXIMUM
SANCA 3 2260 1
2321
56K
3
2314
ACORN 1 D.P. ?
2314
None outside
TSAR
OCS-ADMIN 1 2260 ?
2314
None outside
TSAR
Estimate, TS Applications of 1 July 1970
'CORE
APPLICATION # TERMINALS -TYPE DASD
TYPE
MAXIMUM
SANCA 3 2260 1
2321
56K
3
2314
ACORN 1 D.P. ?
2314
None outside
TSAR
OCS-ADMIN 1 2260 ?
2314
None outside
TSAR
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Attachment "C"
PROJECTED TS APPLICATIONS from ISD
Estimate, TS Applications as of 1 July 1969
1. Application:
Number & Type of Terminals: 2 CRT for query
1 hard copy output
Amount of DA Storage
Required: 1060K bytes
Program Requirement - Core: 100K bytes
2. Application:
Number & Type of Terminals:
Amount of DA Storage
Required:
Program requirement:- Core:
use same terminals as PARIS
1500K bytes
100K bytes
3. Application:
Number & Type of Terminals:
Amount of DA Storage
Required:
Program requirement - Core:
4. Application:
Number & Type of Terminals:
Amount of DA Storage
Required:
12,000K bytes
100K bytes
12,000K bytes
Program requirement - Core: 80K
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 2
5. Application: OSR
Number & Type of Terminals: one 2250
Amount of DA Storage
Required: 150,000K bytes (using Litton
figures)
Program requirement - Core: ?
6. Application: FMSAC
Number & Type of Terminals: one 2250, two 2260, one hard
copy output
Amount of DA Storage
Required: 7,000K bytes
Program requirements - Core: 250K bytes
7. Application: NIPROD
Number & Type of Terminals: one 2260, one hard copy output
Amount of DA Storage
Required: 7,560K bytes
Program requirement - Core: 100K bytes
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
SUBJECT: Report on IBM360/67 Test Using CP
On 21-22 Junp 1968 ====I OCS
STAT withi IBM went to
IBM, , to use e IBM /67 to test
the CP (Control Program) software written jointly by Lincoln
Labs of M.I.T. and IBM. The 13 programs, 12 FORTRAN and 1
ALC, which OCS has used as a benchmark to test scientific
processing capability and another 25 programs of PL/1
compiles, links, and GO's which (with the exception of one
grinder, i.e., CPU bound) were essentially I/O bound types
were used to create a mix which was meant to typify the
OCS environment.
The initial two hours on the machine were ession
of errors, many human and one machine. IBM
had generated virtual machines on a tape an is w ich would
represent the OCS environment. His car had been broken open
and the tapes and disk pack were vandalized. A substitute
CP system which did not possess the recent optimization
features was substituted at the last minute. The destroyed
system was on 2314 disks, but unfortunately the substitute
system used a combination of the 2301 drum which is faster
and the 2311 disks which are slower. The tradeoffs of the
two alternatives are so variable that no attempt will be
made to compare their relative performances except to state
that it appears that they are somewhat equivalent. This
Mod 67 which was used had a very large variety of experimental
components and part of the human problem was to vary the
unwanted components off-line. For example, a new experimental
communications controller was on-line and we had been told to
disconnect it since it degraded line interleaving consider-
ably and caused other problems. In the "virtual" confusion
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 2
we forgot to disconnect it, and after disconnecting it later
after running awhile, one 256K memory box gave a machine
check. At this time we discovered we had used three (3)
memory boxes instead of the desired two (2), and thus took
the opportunity to switch out the suspected defective box
and the communications controller. Sixty (60) terminals
were on-line in the building and some number from external
sources were also connected to the system. Unfortunately,
as we were delayed to a starting time of 2326 on a Friday
night, the users of the terminals no longer called in, and the
terminal exercise had to be performed by the five of us who
were conflictingly busy with the CP and the test measurements.
The final CP environment included a 2301 drum, 2311
disks, tapes, card reader, printer, 60 in-house terminals,
an undisclosed number of external terminals, and 524K of
memory of which 80K was reserved for CP. On this 444K (net)
memory machin( .we created eight (8) virtual 360 machines.
Of these eight (8), because of lack of manpower, we were
able to activate only as many as three (3) machines at one
time. Virtual machine #1 (symbolic name of BARR) was given
524K bytes of core plus various peripheral space and on its
library we put two (2) systems, OS and CMS (Cambridge Monitor
System). Either OS or CMS could be active at one time.
CMS has terminal handling capability and since it uses virtual
memory, it can handle N terminals (of course, limited by
lines and controllers). It also contains FORTRAN-G, SNOBOL,
PL/l, and ALC, all of which can be used in batch or conver-
sational terminal mode.
Virtual Machine #2 (symbolic name of BATCH) was identical
to BARR (#1), i.e., it also used 524K of virtual memory and
contained both OS and CMS.
Virtual Machine #3 (symbolic name of COME) was identical
to the other two but only used 262K of virtual memory. This
one was to be used for our terminal testing, and its CMS
also contained the conversational PL/l compiler.
Thus, in summary, a 524K Mod 67 with a net actual memory
of 444K (80K used for CP) was made to act like three (3) other
machines, two of them with 524K memories and one with 262K
memory. Five (5) additional but non-activated systems were
also included.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 3
Virtual
IBM 360
(BARR)
524K
Virtual
IBM 360
(BATCH)
524K
Mod 360/67
444K + 80K(CP) = 524K
Virtual
IBM 360
(COME)
262K
VM
#4
VM
#5
#6 #7 #8
t
V
ACTIVATED
OS
CMS
CMS (with
conversa-
tional PL/1
1 1 ~~
I I
NOT ACTIVATED
Mix of Jobs Used in Test
Twenty (20) CLG (compile, linkedit, and GO) PL/1 programs
were used. GO times on nineteen (19) of these were very
minimal, in contrast to the compile and linkedit times. One
program, P-44 was specially written as a small PL/1 grinder,
i.e., a PL/1 program which was CPU bound in a tight loop.
One ALC compile, three (3) FORTRAN compiles, and five (5)
FORTRAN CLG's, all of which were test programs used in previous
benchmark measurements and all of which were short were in-
cluded. Another relatively long FORTRAN job, P-70, which was
CPU bound in an infinite and medium size loop (required some
paging) was used to saturate one virtual machine for the final
2-1/2 hours of 3 hour test run. No performance figures can
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 4
be compared or are pertinent for this job. However, it is
valid to state that this steady processing assured a good
measure of a variety of programs running concurrently with
another loaded virtual system/machine.
Programs used on the terminal were FORTRAN, ALC, and
PL/l. The load generated by the terminal users cannot be
measured or compared except to say that the third virtual
system/machine was active. See Charts A, B, and C for
performance and chronological running statistics.
Initializing of Virtual Machines
Initializing of a virtual machine and its system is a
matter of seconds. This appears to be as simple as possible.
Compatibility with HASP
A HASP/OS system is just another system and can be
loaded like any other. In our test, we did not load HASP/OS
but rather just OS. CP does its own spooling and HASP is
not necessary; in fact, a HASP/OS system will SPOOL to the
HASP SPOOL queue and this queue will in turn SPOOL to the
CP SPOOL queue. Obviously, this is redundant and inefficient.
For a shop which wishes to process jobs in CP as background
and yet maintain compatibility with other HASP systems in the
shop, an operator command is available for the operator to
ATTACH a device such as a printer or card reader to the HASP
virtual machine in a "sole use" mode. Thus HASP/OS jobs can
be run under a virtual machine without change if the operator
ATTACHES the HASP devices to that system. This does mean
extra hardware and cost. HASP jobs will run under CP with-
out this feature, but accounting information is not gleaned
from the job card. In fact in our test, this happened several
times when the old HASP job card was inadvertently left in
the deck. (See Chart "A", Jobs PROB45 and PROBPL.)
Efficiency of Background CP Programs
Most programs ran about four (4) times as long in the
background under CP as they did under HASP (See Chart "C").
However, always two jobs were being processed concurrently
while the terminal system was exercised. Thus, it can be
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 5
stated that background capability on a Mod 67 is approximately
one-half of the Mod 65 throughput capability when a minimal
on-line interaction is performed. Degradation of background
upon activation of additional terminals has not been measured.
When activity is "trivial" on 10-30 terminals, a small linear
degradation is evident (from Lincoln Labs trip). Number of
terminals alone is not a valid comparison measurement but
interaction speed, type of compute, type of query, etc. are
all very complex and little understood variables and thus
conclusions on degradation are risky and will not be made.
It is valid to so state that on-line response is excellent
in all cases observed on the previous trip. The system should
be viewed primarily as an on-line system with good handling
of background jobs in the remaining time.
The performance of one program, P-44, is interesting.
It had the highest ratio of HASP time to CP time (18/45) and
it is a grinder. However, it is PL/1 and small. Its rela-
tively efficient performance (it was running concurrently
with another grinder and with the terminal system) might be
attributed to the fact that it was too small to page or to the
fact that it was PL/l. Impressions gleaned from the periodicity
of flashing lights supports the impression that zero paging
is the chief factor. A corollary might be given that since
all other measured programs which were run paged (i.e.,
compilers, linkeditors, and object code exceeded non-paging
parameters), all other jobs were less efficient than need be.
Again, this is speculation on yet another set of unmeasured
variables. Tentative conclusions to be drawn are:
1. the less paging, the more efficient
2. programs can be written to be paged less and
thus can be made more efficient.
Optimizing Program Efficiency for Background Jobs
Programs can be made to run significantly more effi-
ciently by "packaging." None of these programs was packaged;
however, P-44 inherently (accidentally programmed that way)
was equivalent to a "packaged" program. Thus the 18/45
(HASP/CP) performance of P-44 very possibly is an optimum
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 6
efficiency of the present CP version using two background
systems.
Packaging is done by a special routine at the time
programs are ready for production. The new program is
executed under "PACKAGING" control and a paging optimization
chart is generated. The programmer using this chart now re-
packages his source deck before final compile. For overall
shop efficiency it is extremely desirable to make this extra
pass. Performance increases are very significant.
Jobs which run poorly
Jobs which are not modular or which have long series
of in-line code can rarely be packaged efficiently. Large
matrix programs would be extremes of inefficiency and pack-
aging would help very little.
Significance of the Wait Light
On most IBM 360's the wait light is "on" most of the
time while processing. Thus the CPU is idle. It is often
stated that the more the wait light is off, the more efficient
the system is and "turning off the wait light" many times be-
comes a system objective. Generally, this is a valid objective;
however,?this objective' should be qualified in that only is it
valid when the CPU is doing productive work. For example,
the CPU can be tied up in "system overhead" and this is a
good thing only in whether or not the net throughput is
better.
For what it is worth, the wait light was mostly "on"
until P-70 was loaded. This grinder turned the light off
and the final 2-1/2 hours were run with the light off. Thus,
it can be said that a grinder in any of the virtual machines
will absorb all surplus CPU power. Also, it is suggested
that if several non-CPU bound virtual systems are processing
concurrently and if the wait light is "on", it is advantageous
to load systems until the light is off. One can speculate
on the implications of this, such as "What if a job becomes
progressively more of a grinderas it processes"? "Is it
valid to process a 20 minute job, - sat for four (4) hours
because the processing power is surplus and thus is free"?
In any case, it can be stated that the CPU can be saturated
so that no CPU power is wasted; however, it must be recognized
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 7
that 10-45% of this power is used in overhead and not in
productive work for the customer programmer.
Utility Work in the Background
At one point in the test (See Chart "A") it was re-
quired to scratch some data sets on disks. For six (6)
minutes one virtual machine was used for this purpose at no
loss in throughput on the other virtual machines. Operators
will appreciate this capability.
At the very end of the test while the infinite grinder,
P-70, was processing, another customer had priority over
us and requested the machine. We processed his programs
which were tape and disk utilities and which were I/O
dominant type and he went away happy even though our pro-
grams were not interrupted. His programs which were small
utilities (I/O and little or no paging) seemingly ran at
full speed.
OS versus CMS
FORTRAN-G, PL/l, and ALC (plus others) are available
under CMS. OS pages excessively and if programs are compiled
under CMS rather than under OS, performance is significantly
increased. All compilations in the test were under OS and
thus all performance measurements are the worse case.
It is sufficient to state that improvements are avail-
able by using CMS rather than OS for certain jobs. In-
compatibilities are reported non-existent or minimal, but
until actual comparisons are made, reservations are in order.
Human Factor in Comparison of HASP and CP
Previously in this report, it was stated that the back-
ground of CP had one-half the throughput power of HASP on
Mod 65. (See Chart "C") This statement and the times shown
are extremely biased in favor of HASP. In the HASP test
whic was run under conditions controlled by the tester rather
than by the computer operator, the CPU and/or the I/O was
active 100% of the time. Measurements of actual computer
performance in OCS during prime shift show that the computers
are totally idle at least one-third of the time in that the
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Attachment D
Page 8 (Revised)
machine is waiting for the operator to do something and that
another significant amount of time is used for I/O during which
time the CPU is idle. HASP SPOOL time would be included in
the latter figure and when HASP SPOOL is the only I/O, the
computer is also waiting for the operator to do something. The
latter case could not be measured, but observations show that it
is significant. The above measurements were taken on prime
shift when supervision of operations should be optimal and when
the percentage of non-set up jobs to set up jobs is highest which
is also an optimal condition. All of this suggests that the usual
effective usage of the Mod 65's is significantly less than the
figures given in Chart "C".
Stating a conclusion from the above observations that the
Mod 65's are usually only 50% effective because of human
factors would invite arguments as to whether the operator and
human procedural degradation is 20, 30, 40, 50, or 60%. In
any case, it is significant, and in our comparison test case, it
was 0%. If we assume that this human degradation of actual
throughput is 50% in a 24-hour day (which appears reasonable
from the above), and if we assume that an operator can keep at
least one virtual machine active at all times, then the comparison
performance figures will show that background processing capability
of CP-67 is equivalent to the current effective throughput of an OCS
Mod 65.
Thus, now background processing capability is at least 50%
of a Mod 65 and effective throughput will range from 50 -100% depending
upon human and procedural factors.
The performance of the Mod 67 and the CP software was
impressive. Previously, we had seen it perform terminal work
with rapid response time and the objective of this test was to
measure background capability. Obviously, the OS running under
CP is compatible to OS on a standard 360. Reliability during the
(3) hour test was perfect, even though a core box had a process check
during the preliminary test time. Background throughput was measured
and in the sense that this is an extra benefit, it is good.
A 444K machine which handles eight (8) virtual machines,
three of them active each with two operating systems and two
of these active with larger memories than the Mod 67 is
impressive. Also, it is difficult to reduce the large number
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Page 9
of variables to specific measurable conclusions. Meditation
on these variables generates an infinite number of hypotheses
which could be tested and obviously this alone is a strong
recommendation. The system is the most flexible on the
market and it can absorb or test nearly any of the many
publicized IBM/360 time sharing systems available. For
example, the OCS TSMON system could be run as a separate
system on a virtual machine concurrently with other systems
being tested. Even MVT (Multiple-variable-tasking) OS
will run as a system on a separate virtual machine.
Much has been learned about time sharing over the past
several years and many former notions have been discarded.
It does take much raw CPU power to drive a large time
sharing system satisfactorily. This power is necessary to
provide excellent service at all times, even though a peak
of on-line interaction may be occurring. Such power is
expensive and it makes economy sense to use that power during
the troughs of on-line interaction to do background process-
ing. The Mod 67 has the power to do the above processing
concurrently.
STAT
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Chart "A"
PROCESSING LOG from CP TEST
JOB Time Time Elapsed Virtual
in out Time Machine
L3D406 1126 1130 4 Batch
PROB09 1127 1132 5 BARR
L3D400 1130 1131 1 BATCH
L3D40A 1131 1141 10 BATCH
PROB10 1132 1138 6 BARR
PROB11 1138 1142 4 BARR
PROB12 1142 1146 4 BARR
-- 1141 1144 3 BATCH
L3D405 1144 1150 6 BATCH
PROB15 1146 1150 4 BARR
PROB13 1150 1154 4 BARR
L3D440 1150 1157 7 BATCH
PROB14 1154 1158 4 BARR
PROB-70 1157 -- - BATCH
PROB15A 1158 0004 6 BARR
PROB17 0004 0009 5 BARR
TERMINAL 0005 0104 59 COME
PROB18 0009 0012 3 BARR
PROB37 0012 0017 5 BARR
PROB31 0015 0025 8 BARR
PROB33 0025 0030 5 BARR
PROB44 0030 0115 45 BARR
L3D270 0115 0126
TERMINAL 0117 0205
PROB51 0126 0128
PROMMA 0128 0133
PROBPL 0133 0133
11 BARR
48 COME
2 BARR
5 BARR
0 BARR
L3J265 0133 0138 5 BARR
UTILITY 0138 0144 6 BARR
PROB08 0144 0148 4
L3D404 0148 0158 10
L3D403 0158 0200 2
L3D407 0200 0206 6
L3D402 0206 0219 13
PROB01 0219 0223 4
UTILITY 0226 0240 14
BARR
BARR
BARR
BARR
BARR
BARR
BARR
FORTRAN
PL/1 Job interrupted to set v
disks (1 minute?)
FORTRAN or ALC
ALC or FORTRAN
PL/1
PL/1
PL/1
not used, operator behind
FORTRAN
PL/1
PL/1
FORTRAN
PL/1
FORTRAN, program was a grinde
in an infinite loop and at
0243 was terminated.
PL/1
PL/1
FORTRAN, ALC, PL/1 programs
run by various customers.
PL/1
PL/1
PL/1
PL/l
PL/l, was a grinder and backe
to grinder 63D451
Dupe JOB card, ignored, same
as L3J270
PL/1
PL/1, FORTRAN, etc.on termina
PL/1
PL/l, short grinder
Dupe JOB card, ignored, same
L3J265
PL/1
Vir Machine used for 6 minute
for utility work
PL/l, JCL error because of
change in system. Not sub-
sequently run.
PL/1
FORTRAN
FORTRAN
FORTRAN
FORTRAN
PL/l
Various disk & tape utilities
were run for other customer
Approved For Release 2005/06/03 : CIA-RDP78-0394856M0fjq at 0205
BARR at 0223
BATCH " at 0243
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Chart "B"
PROCESSING LOG from HASP TEST
HASP Job Name CP Job name
Time (minutes)
PROB #1
PROBO1
98
SCHOOL PROBO8
.
1
40
EXERCIZE PROB09
START
PROB10
START
PROB11
.
1.16
1.35
1
16
PROB #12 PROB12
.
0
90
PROB #15 PROB15
GP2
PROB13
A
PROB14
A
PROB15A
PROB #17
PROB17
.
0.92
0.96
1.01
1.15
0
94
PROB #18 PROB18
P
.
0
88
ROB #37 PROB37
.
RANDOM
PROB31
MAJIC33
PROB33
MAJIC44
PROB44
BIN-TEST
PROB45
CLOCK
PROB51
MALTA
1.16
1.87
1.07
18.34
1.08
0.99
PROBMA
PLOT
PROBPL
1.35
1'. 34
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Chart IT
TIMING COMPARISONS of HASP 'versus BACKGROUND a on CP
Job
Type
(Minutes)
HASP
Mod 65
b
CP-Background
400
ALC
C
0.25
1
40A
FORTRAN
C
2.46
10
402
FORTRAN
CLG
2.70
13
403
FORTRAN
C
0.31
2
404
FORTRAN
CLG
2.79
10
405
FORTRAN
C
1.34
6
406
FORTRAN
CLG
1.49
4
407
FORTRAN
CLG
1.40
6
440
FORTRAN
CLG
1.98
7
P-1
PL/1
CLG
.98
4
P-8
PL/1
CLG
1.40
4
P-9
11
If
1.16
5
P-10
if
if
1.35
6
P-11
"
1.16
4
P-12
.90
4
P-13
.96
4
P-14
1.01
4
P-15
.92
4
P-15A
1.15
6
P-17
.94
5
P-18
.88
3
P-31
1.87
8
P-33
"
1.07
5
P-37
1.16
5
P-44
"
18.34
45
P-45
1.08
11
P-51
0.99
2
P-MA
1.35
5
P-PL
1.34
5
54.73
198
a. Foreground job consisted of more than 60 terminals on-line
but during test only one was active on which was processed
conversational PL/l, FORTRAN, and ALC. No measurement
possible for foreground tasks.
b. Two jobs were running concurrently in the background. Thus
an effective measure of background power suggests that all CP
times could be halved for comparison purposes.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
HARDWARE SYSTEMS REJECTED UPON PRELIMINARY INVESTIGATION
A number of time-sharing systems other than IBM's were investigated
(CDC, GE, UNIVAC, SDS, DEC, RCA) but were not seriously considered
primarily because they were not compatible to the Computer Center pro-
duction systems (IBM 360/65) and could not be used as back-up to the batch
workload. In addition to the compatibility problems none of these systems
could completely satisfy our TS processing objectives. Some of these
systems were not operational yet and others were felt to be lacking in one
or more of the following areas:
CPU processing power
Adequate direct access storage devices on-line
Query language
Terminal response time
Information retrieval package
Software security
Shared data sets
Scheduling algorithm
CRT support
SDS - 930/940 Supports 16 terminals, response poor, lack of
adequate on-line storage capacity, no query
language or IR package.
PDP - 6 JOSS type language only.
GE - 645 MULTICS - not completely operational, perfor-
mance poor.
GE - 265/235 BASIC only.
GE - 635 Time sharing system to be implemented under
GECOS - not operational yet.
CDC - 6000 RESPOND system editing and remote batch capa-
bilities only.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
COMPARISON OF HARDWARE COSTS
(In Dollars)
Model-50 Model-65 Model-67
CPU 1052-7, 512K byte
core, 2 selector channels
1 max channel (models 65
and 67 include 1 sort
selector channel)
Drum (2820 control unit -
2301 drum)
LCS (one million bytes)
Peripherals
1 Current system
TSMON TSMON
TSMONI RUSH ADEPT TSMONI RUSH ADEPT CP/CMS
22,290 22,290 22,2902 38,692 38,692 38,6922 42,682
4,620 4,620 4,620
6,500 6,500
19, 679. 50 19, 679. 50 19, 679. 50 19, 679. 50 19, 679. 50 19, 679. 50 19, 679.50
41, 969. 50 48,469. 50 46, 589. 50 58, 371.50 64, 871.50 62, 991.50 66, 981.50
2 Additional cost for RPQ on this system not included here.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
ATTACHMENT G
Price of RCA 70/46
Processor (262K)
$16,446
Peripheral Units
.4, 705
Random Access
12, 253
Communications Devices
3, 876
$37,280
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
ATTACHMENT H
IBM 360/67
SYSTEM CONFIGURATION
Lease per Month Purchase
2067-1 Processing Unit
17,585
711,490
#4434 Floating Storage Addressing
25
1,100
#7920 1052 Adapter
232
10,545
2365-2 Storage Unit
2
@ 9,530
19,060
795,400
#7123 7 Bit Storage Protect
N.C.
#8035 2067 2067 Attachment
N. C.
1052-7 Console I/O Keyboard
65
2,725
2860-2 Selector Channel
3,090
143,750
#9502 1st Channel on Buss
N. C.
2870-1 Multiplexor Channel
2,265
106,700
#6990 Selector Subchannel
410
17,940
2314-1 Direct Access Storage Facility
5,410
244,440
2820-1 Storage Control
2,370
108,930
2301-1 Drum Storage
2,250
96,000
2821-1 Control Unit
1,000
45,100
#3615 1100 LPM Printer Adapter
77
2,910
2540-1 Card Read Punch
680
33,950
1403-N1 Printer
900
41,200
#8640 Universal Char Set - Mod 3 W1 10
450
1416-1 Inter Changeable Train Cartridge 100
3,000
2701-1 Data Adapter Unit 206
9,410
#3815 Expanded Capability Feature 25
1,200
#3855 Expansion Feature 82
3,640
RPQ F1804A M25697 CK Polynomial Qty 2 90
3,600
RPQ F18049 M24802 Intr Processor Comm Adapter Qty 2 480
19,000
RPQ F1804B 816190 Full Duplex 188B Interface 118
4,130
2848-3 Display Control
435
18,530
#4787 Line Addressing
10
450
#5340 Non-Destructive Cursor
10
430
#3859 Expansion Unit
46
1,890
#5341 Non-Destructive Cursor Adapter 4 @ 5
20
860
#3357 Display Adapter For Mod 3 4 @ 103
412
15,520
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
-2-
2260-1 Display Station
31 970
2841-1 Storage Control Unit 540 26,430
#4385 Scan 36 1,360
#6118 Record Overflow 10 400
#8079 2321 Attachment 180 6,790
2321-1 Data Cell Drive 2,885 132,400
2803-1 Tape Control 670 31,620
2401-3 Tape Units 3 @ 810 2,430 110,280
2702-1 Transmission Control 875 39,580
#4615 IBM Terminal Control Type I 36 1,575
#3233 Data Set Line Adapter 20 950
#3853 Expansion Base N. C.
#7912 Telegraph Term Control Tpy 11 36 1,575
#4635 IBM Line Adapter 4W Limited 5 @ 23 115 5,425
#8055 2741 Break 10 450
2741-1 Communication Terminal 82 3,100
#4635 Line Adapter 3W 3 135
#4708 Interrupt 2.50 115
$65, 424. 50 $2, 807, 445
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
SYSTEM CONFIGURATION
2920
Approved For Rel~e 5/06/03: CIA-RDP78-039
2702
? Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Description of IBM 360/67 and CP/CMS Software
The IBM System 360/67 was specifically designed to overcome the
problems encountered in programming and operating a time-sharing
application. The following features are considered necessary in an
on-line computing environment:
1. CPU Speed - The 200 .anosec internal cycle time assures
a fast response to console users. The necessity of a high-speed
CPU in this environment is not to handle a few staggered demands
for service, but to insure that when multiple instantaneous user's
demands exist, the apparent terminal response is not degraded.
Most computer systems designs are built around a turnaround
criteria of hours; and job load averages in this spectrum of time
can be analyzed to determine the best CPU cost vs. time tradeoff.
In a time-sharing system this same system design technique is
invalid. The on-line system which uses only response as a measure
of its acceptability requires a high-powered CPU, a CPU which is
excessive in regard to the computing demands of the on-line users.
Any inefficiency in this type operation suggests that the system also
accommodate another class of user, the background job. This low
priority work must be available to absorb the excess CPU capability.
2. Dynamic Relocation - This feature on the Model 67 insures
the security of user data and programs which co-exist in the same
real memory. It also provides the control programs with the hardware
necessary to overcome the core fragmentation problem; therefore,
only programs and data which are actually in use require residence
in high-speed memory.
3. Channel Architecture - The channels in Model 67 are self-
contained; that is, they are not part of the CPU as they are on the
less powerful models of the System/360. This arrangement has two
advantages--first, a hardware error in the channel does not bring
the system down completely; second, when this architecture is further
extended by the inclusion of a channel controller IBM 2846 in the
system configuration, the probability of interference between the
channel and CPU memory demands is reduced.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
2
4. Reconfiguration - In the Model 67 system equipped with
a configuration console, the probability of extended system
outage is considerably reduced. The operator need only switch
out defective components and gracefully degrade system performance.
The real advantage of any computing system is, of course, its
software. In this respect, the System 360/67 is unique. The CP/67
which is the time-sharing part of the system program has evolved
from an earlier system CP/40 which was operational over two years
ago.
The CMS (Cambridge Monitor System), which is one of the on-line
programming systems available to the terminal user, is the same
system that operated with CP/40 and therefore has benefited through
two years of use.
The specific advantages of the CP/67 come from its creation in
a dynamic sense of many "virtual computers." This technique allows
the terminal user to regard his console as an operator's console,
1052, and he programs as if he is running on a 360 computer by
himself. This organization permits the individual terminal user a
multitude of already existant programs and lowers the cost of converting
to an on-line programming environment.
The CP/67 system has the following attributes:
1. Compatibility - Most significantly by being able to run OS,
it is completely compatible with the rest of the installation.
2. Openended - CP/67 allows execution in a virtual machine any
360 programs which are not timing dependent nor have data driven
1/0.
3. Security - The virtual machines are accessed through a
password scheme and since all memory and I/O references invoke
mapping by hardware and software, the security of the individual's
data is nearly absolute.
4. Maintainability - The system is easily maintained for two
reasons; first, because it is relatively small; second, because it is
very modular and the individual modules have very little dependence
on one another.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
5. Proven System - The system has been over two years in
operational development and has been successful in operation
in a customer site (Lincoln Lab) for over a year where it
currently supports 30 terminals.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Questions and Answers on Model 67 and CP
1. What is CP?
A Time-Sharing System for the Model 360/67 hardware.
20 What does it do?
This software system, working in conjunction with the hardware,
creates a unique environment called the virtual machine.
3. How does it compare with MFT and MVT?
There is a functional difference between CP and other current
operating systems. In CP, the common user functions usually
called data management are separate from those functions
necessary to perform multi-programming. Under the other
systems, multi-programming and data management are combined
and usually user directed.
4. Under CP on the Model 67, how many virtual systems can be "sysgenned"
into the system?
This is limited by the amount of space, I/O, and secondary storage
which are available.
5. What is a virtual system?
Virtual system is a simulated 360 environment in which all inter-
user conflicts are resolved through mapping by either software of
hardware, or both.
6. What systems can be operable under CP? (OS, CMS? ? ? )
Any non-timing dependent/ 360 program is operable in a virtual
system,
7. What is the limit to the number of users on line at the same time
under CP?
The limit on the number of users depends on the load conditions
but currently there are 30 on-line users at the Lincoln Laboratory.
Plans call for another control unit and then the number will double.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
8. When are the bounds of each virtual system defined? Can they
be altered at any time?
The bounds of each virtual system are defined in the user
directory. It can be altered by creating a new directory
before start-up.
9. What device support is available under CP?
A.
B.
C.
D.
E.
F.
G.
H.
2260
2250
2741
Disk drives 2314, 2311
Drums 2301, 2303
Large core storage is being
Printers --No restriction as
implemented at Washington State.
to the number of printers.
Card Reader-Punch--no restriction as to number.
10. Can one use RJE's with this system?
Yes, current plans include 1130 support as an RJE terminal.
11. Would it be practical?
12. Is there much degradation to the system if they are all on at once?
Degradation in a time-sharing paging environment is a function
of the load of the terminals. In a correctly operating system,
degradation should be linear as the load.
13. Can any one virtual system interfere with any other virtual system?
Even deliberately?
14. If data is written on a disk or other I/O device by one virtual system,
can it easily be accessed by another virtual system?
Access to I/O data is only achievable through the directory.
Therefore, users can be prohibited or permitted access to one
another's data by altering the directory before start-up.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
3
15, Can remote terminals be used with this system?
Yes
16. Can the remote terminals be on-line with the batch?
17. Is there any chance that one remote terminal may retrieve
data from another area other than the area he is using?
18. Does the remote terminal degrade much from batch performance?
Even with a compute bound job?
Degradation in the time-sharing system is severe for jobs which
misuse core even in a batch system. Jobs which are written
to run well in the batch environment tend to perform well in
the time -sharing environment.
19. What is CMS?
CMS is the Cambridge Monitor System. It is the Operating System
used by the terminal user. It provides on-line capabilities such
as FORTRAN, PL/l, context editing, SNO BOL.
20. How does it compare with OS?
In comparing CMS with OS, with CMS we have a subset of the OS
data management functions and a subset of the languages available
under OS and no multiprogramming facilities. CMS is comparable
to FMS (Fortran Monitor System) for the IBM 7090,
21. Which types of jobs, I/O or compute bound, run best under the CP
system?
I/O bound jobs tend to run best under the CP system.
22, Why?
Because there is a greater amount of time available for multi-
programming with these type jobs.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
4
23. Is there a significant increase in performance with faster data
rate I/O (2311 vs. 2314 vs. 2303 vs. large core)?
System performance with regard to the I/O system depends
heavily on the ratio of core space and CPU demands to I/O
demands.
24. With only 444k bytes of physical core available, how can the
machine act as if it has 1 meg. bytes available?
By the use of a hardware relocation device which maps all
storage references, those pages (i, e. , 4096 bytes) which are
not in core are brought in from secondary storage and the
relocation device is updated.
25. What is paging?
Paging is the artificial division of a user program's address
space. In the 360/67, this space is divided into 4096 byte
pages. Addresses generated by the CPU pass through and
are acted upon by a hardware relocation device before reaching
the execution store. Those pages which are not in the physical
core store are retrieved from secondary storage devices (drum,
disk). The entries for the relocation device are changed to
reflect the swap from drum to core.
26. Is only part of the program stored in core at any one time?
The part of the program required for execution is in core.
27. If so, where is the rest of it?
On some secondary storage device (drum or disk).
28. If two small programs are operating under two virtual systems,
might everything remain in core from both systems at the same
time ?
Yes, if the sum of the two active page requirements is less than
or equal to the number of available pages in a physical core.
29, Under CP and the concept of virtual systems, how large can a
program be before it runs out of core?
16 million bytes, i. e. , the number of bits in the address field.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
30, If one virtual OS systems fail, does it bring down the whole
system?
No
31. If CP fails, does it bring down the whole system?
Yes
32, Does CP have both input and output spooling?
Yes
33. How can the output from separate jobs be kept separate?
By prefixing the printed output with a user ID,
34. Is there a way to direct certain job output stream to a certain
printer ?
This facility is currently being implemented.
35. From the same system input device (i, e., card reader-punch)
how does the input go to the intended OS virtual system?
By preceding the user input cards with a user ID card.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03
W8A000100090001-8
ATTACHMENT I
26 April 1968
MEMORANDUM FOR: Computer Science Advisor
SUBJECT : Planning for IBM 360/50 Replacement
1. I would like you to form an ad hoc group within the Office to
study the alternatives and recommend a course of action regarding the
replacement of the 360/50 in the Computer Center. As you know, this
computer will be moved to the DD/P in January 1969.
2. In analyzing the alternatives, the following factors should be
considered:
a. The system will be used principally in the time sharing
mode. Background processing for efficient computer utilization
is desirable but not mandatory.
b. I do not believe we can plan to move into a completely
integrated software system including multiprocessing, multipro-
gramming, and time sharing as the standard Operating System in
the Center. Compatibility with the software and hardware used in
the Center is certainly desirable, particularly if main frame con-
nections are feasible for passing tasks back and forth.
c. We should assume that by January 1969 the customer re-
quirements and our experience with time sharing will be such that
a stand alone production time sharing environment will be feasible
and desirable.
d. Movement toward a production environment will make sys-
tem experimentation and extension more difficult than it is now, but
through judicious planning (and perhaps the use of IPRD facilities),
we should be able to continue experimentation at a reasonable pace.
e. Costs must stay within current budget estimates. Supple-
mentary funding could be justified only if major new requirements
were surfaced.
Approved For Release 2005/06/03: CIA-RDP78-
T LEA
1948AO00100090001-8
II~3'J l~C i e ..
!rw?vu,.uu ~~ ~
Approved For Release 2005/06C-ZP78-03948A000100090001-8
f. Conversion to the new equipment should be as painless as
possible.
g. We have no firm basis for projecting load other than what
we know about existing applications and informal requirements.
We should assume that the known requirements will constitute the
minimum load to be expected.
3. The above factors tend to reduce the number of hardware alter-
natives that must be considered seriously. In my view, the following are
the more obvious ones (in no particular order, with no comment):
a. A 360/50 starting with the configuration identical with the
current one but expanding as needs arise (LCS, AMCS, 2314's,
etc. ).
b. Two stand alone 360/40's, one for Agency sensitive data,
the other for everything else.
c. A 360/67 using a minimal configuration needed for TSMON
(with or without CP67).
d. A 360/65 configured for time sharing.
e. A Spectra 70/46.
4. Software alternatives are perhaps the more difficult to evaluate.
Some random thoughts: We have to assume that the programming talent
that can be applied to time sharing software and related application serv-
ices (in quality and quantity) will always be less than optimal. But I
think the "shoe string" effort thus far can be expanded over the next year
because we have encouraging tangible results. We should concentrate on
building and expanding services peculiar to our installation, perhaps at
the expense of building monitors or spending time making them more
efficient. As the system becomes saturated, our first question should be
whether modest increases in hardware would keep the system going rather
than to immediately task our available programming manpower to squeeze
more from the existing hardware. Only when serious overload is expected
should we look to improved or new monitors. In this way, I would hope
that we could keep the system going until efficient monitors that meet
our needs become available from the outside, using our people in the
meantime to build more and better application services.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
SIC R1ET
Approved For Release 2005/06/0 -03948A000100090001-8
5. Your findings should be available by 1 June so that a decision
can be made and an order placed with sufficient lead time.
25X1A
I
Deputy Director of Computer Services
cc: D/OCS
TRC Members
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
"z-CI
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
26 June 1968
ATTACHMENT J
SUBJECT: Trip to RCA on Spectra 70/46
1. On 17 June 1968 ent to Cherry STAT
Hill, New Jersey, to test the Spectra 70/46 and talk to systems personnel.
(See attached for a description of Spectra 70/46 as written by Totaro of
Auerbach and distributed by RCA.)
2. Meinstein, RCA, stated RCA's time sharing objectives as
follows:
a. "To provide concurrent service to a large number of users
at remote points. "
b. "To supply the user with immediate access to a computer
so that it appears to him that he is the only user. "
He gave an overview description of the hardware which is included in the
attachment. He explained that 524K of memory was rejected because of
cost and thus the 5-46 was a 262 byte machine. System design programs
are separated into two classes.
Class 1 - that which is brought over, must be loaded in entirety,
must be loaded contiguously, must use private volumes, must never
be paged, and memory is not to be relinquished.
Class 2 - Program is created on S-46 with TDOS compilers,
pages are pageable, certain pages are not to be moved, only the
page with an entry point must be loaded to start, may share public
and/or private volumes.
In data management, SAM and BDAM are available on all devices except
RACE. BTAM (Basic Tape Access Method) is available. ISAM is avail-
able on all DASD except on RACE.
All code is re-entrant. Is this because of insufficient memory size?
Is it worthwhile? This was rather unusual since there is a trend to forget
re-entrant code on paging hardware. Cataloging is available by name.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
JCL and TCL (Terminal Control Language) are a single language
which is an excellent feature. The operator can designate number of
pages limited to a class l program.
3. Performance Tests. In order to demonstrate that the S-46 works,
an unofficial demonstration was given. One typewriter terminal was
activated and then 39 copies of another program were simulated by attach-
ing a S-45 to the S-46. Performance and response were impressive.
However, as most time sharing implementers have discovered, a syn-
chronous and perfectly interleaved load is not the same as humans
asynchronously interacting with a system. Whether RCA really has a
better time sharing technique than others could not be determined. If a
machine as low powered and with as small memory as the S-46 can handle
40 users with 1-2 second response time, RCA has a winner. In conclusion,
I am willing to go on record with my personal opinion that RCA is in for a
rude awakening on performance. In my opinion, RCA on the S-46 will not
provide 40-48 users with adequate response time. When load peaks and
especially with c complicated queries or calculations, I believe the
response time will degrade to many seconds, or even minutes and cus-
tomers will be dissatisfied. Unfortunately, RCA's system cannot be
adequately exercised and measured.
4. Device support.
Data Cell. RCA offers its RACE; and even though direct access
is not supported under S-46, they offered to help move SANCA over
to the RACE.
Fast Direct Storage. RCA will not have its own large disk until
1970, but they offered to interface with a 2314.
2250. They can interface through an 1130. They have no com-
parable piece of equipment manufactured by RCA.
5. Language Support. RCA supports FORTRAN and COBOL. The
statement was made that a S-46 could handle jobs as fast as the Mod 65,
even in FORTRAN. OCS is supplying them with the 13 jobs from the
benchmark test so that they can compare their times with a Mod 65.
Several similar statements were very disconcerting and reminded me of
performance claims for the S-70/45 before delivery. However, if what
they are claiming is true (or even partially), OCS should immediately
substitute RCA equipment for all IBM gear. Just to satisfy all parties
concerned, RCA must complete the comparison testing of the FORTRAN
programs.
2
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
6. Summary. Very interesting. RCA has very impressive tech-
nicians. They are making many ambitious claims. If these are really
true, they should be given strong consideration in new time sharing
equipment selection.
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
CPYRGH
T
AUERBACH ON COMPUTER TECHNOLOGY This prorile of RCA's new time sharing sys-
tem and other recent developments in the
Spectra 70 line is extracted from a new 250-
page analysis of the RCA Spectra 70 hard-
ware and software in AUERBACH Standard
EDP Reports, an analytical reference service
published by AUERBACH Info, Inc., of Phila-
delphia.
time sharing
spectra 70 style
By J. BURT '(OTARO
Associate Editor, AUERBACH Standard EDP Reports
AUERBACH CORPORATION
Spectra 70 se-
ries of comput-
ers was an-
nounced in De-
cember 1964,
RCA entered
head - to - had
Recognizing these problem
areas, RCA quietly set about clos-
ing the "facilities gap" between it-
self and IBM. During the past year
RCA has added new and impres-
sive peripheral units to its Spectra
70 line and has greatly increased
the capacity and scope of its soft-
ware systems. Most recently, on
May 4, 1967, RCA announced the
long-rumored Spectra 70/46 Time
Sharing System, a development
that serves to plug the only re-
maining gap of any significance
in RCA's battle to match the proc-
essing capabilities of IBM in the
small-to-medium-scale computer
market.
e
competition
with IBM and
its System 360. From the begin-
ning, RCA promised to provide
more computing power per dollar
than IBM.
Spectra 70 indeed offered much:
a family of five generally compati-
ble computer systems, a high de-
gree of source and object program
compatibility with the IBM Sys-
tem 360, extensive use of mono-
lithic integrated circuitry, three
levels of integrated operated sys-,
tems, and an extensive line of
peripheral equipment.* But the
System 360 offered more-a great-
er variety of hardware and, espe-
cially, more "third generation"
software facilities, such as ran-
dom-access-oriented operating sys-
nurce-language control of
T2ra~,r.: :~cess devices, automatic
data management systems, com-
prehensive data communications
control systems, and support of
time-shared processing.
?For a detnilcd review of the Spectra 70 fam-
ily of computer ayetems, ace "RCA Spectra 70
-A Compatible Competitor," by J. 11. Totaro,
Data Processing Magazine. June 1966. pages
64-61.
Spectra 70/46
The Spectra 70/46 Time Shar-
ing System includes a new Spectra
70/46 central processor, a new 70/
567 high-speed magnetic drum
unit, and a specially-designed soft-
ware system called the Time Shar-
ing Operating System (TSOS).'
Both hardware and software for
the Spectra 70 Time Sharing Sys-
tem are scheduled for first delivery
during the third quarter of 1968.
The 70/46 marks RCA's entry
into the time-sharing computer
market, joining such competition
as the GE-645, IBM System 360/
Model 67, and SDS Sigma 7 com-
puter systems. However, RCA has
designed its time-sharing system
with more modest goals than those
of GE and IBM. RCA's apparent
intention is to remain competitive
in all areas of the small-to-me-
dium scale commercial computer
market without enduring the frus-
trations of the more ambitious pio-
neers in the large-scale commercial
time-sharing business.
The monthly rental of an RCA
Spectra 70/46 Processor with 262,-
144 bytes of core storage is $14,-
125. RCA estimates that typical
70/46 system configurations will
rent for between $25,000 and $30,-
000 per month. Contributing to
the relatively modest prices of
planned Spectra 70/46 systems are
the facts: (1) that the 262K-byte
70/46 Processor is basically an ex-
panded version of the $11,125-
per-month Spectra 70/45 262K-
byte Processor, (2) that only sin-
gle-processor systems have been
announced to date, and (3) that
a maximum of 48 on-line remote
terminal units can be controlled
by the system.
According to its design - goals,
the Spectra 70/46 Time Sharing
System will be an efficient batch
processing system with advanced
multiprograming capabilities; re-
mote, conversational time-shared
operations in time-sliced mode will
be a powerful available facility
that may, in some installations,
consume only a limited amount of
the system's total processing ca-
pacity. The more expensive GE-
Approved For Release 2005/06/03 R~e~$c411~6t49~~~$rA~~zino July, 1967
CPYRGH
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
63.1S and HIM System 360/(Model
67 systems, by contrast, are de-
signed primarily for time-shared
processing.
Noteworthy features of the 70/
46 Time Sharing System include:
? Hardware logic within the
processor to facilitate program
paging and segmentation, provid-
ing simultaneous system access
for up to 48 users at remote ter-
minals.
? Up to 2,097,152 bytes of vir-
tual memory available to program-
ers.
? A fast associative memory for
translation of all virtual memory
addresses to real core storage ad-
dresses.
? A conversational FORTRAN
IV compiler with interpretive exe-
cution capability.
? Full compatibility with the
systems programs and language
processors of the RCA Spectra 70
general-purpose Tape Operating
System (TOS) and Tape/Disc Op-
erating System (TDOS), both of
which are used with Spectra 70/
35, 70/45, and 70/55 computer
systems.
Configurations
Spectra 70/46 system configura-
tions (and their costs) will closely
approximate those of Spectra 70/
45 systems. Any Spectra'70 periph-
eral device that can be used with
the 70/45 can also be used with
the 70/46 (and in generally the
same numbers). The few cnnflgii-
ratio i differences include the 70/
46's required use of a 70/567
Drum Memory Unit and, if remote
terminals are expected to be used
on-line, a multi-line communica-
tions controller. Spectra 70/46
systems are currently restricted to
the. use of a single processor and
262K bytes of core storage.
In addition to the 70/567 Drum
Memory Unit, a minimum Spectra
70/46 configuration will include
at least two 70/564 Disk Storage
Units and two magnetic tape units
for use by the system's software.
The 70/46 Time Sharing System
is designed to take advantage of
the processing efficiencies made
possible through the use of a hier-
archy of system storage devices
that range from magnetic tape
units to disks, drums, and finally
to main core memory.
Storage
Spectra 70/46 magnetic core
memory is available in a single
block of 262,144 bytes. Memory
cycle time is 1.44 microseconds
per 2-byte access. Each byte con-
sists of eight data bits and one
parity bit. Bytes can be grouped
into 4-byte, 32-bit binary words.
This data structure is identical
with that of the Spectra 70/35,
70/45, and 70/55 systems (and,
therefore, with the IBM System
360).
A major feature of the Spectra
70/46 I'rocessot' itt Its read-Duly
control memory unit. This unit
has a memory cycle time of 480
nanoseconds per 54-bit access.
Read-only memory (110M) is pro-
vided in three banks, each of which
contains 2,048 54-bit words (or
"Elementary Operations"). The
Spectra 70/45 Processor, by con-
trast, provides only a single bank
of read-only memory, which is
used exclusively for the emulation
of other computers, such as the
IBM 1401 or RCA 301.
Spectra 70/46 systems cannot
use the emulation technique. Read-
only memory in 70/46 systems is
instead used to implement several
features that facilitate time-shared
operation. The first bank of read-
only memory is used to provide
the microprogramed routines that
will permit the 70/46 to operate
in the Spectra 70/45 processing
mode. The second bank is used to
implement an address translation
table (in a section called the
Translation Memory), an interval
timer, and other hardware features
used by the 70/46 in its paging
and time-sharing operations. A
portion of the second bank and all
of the third bank are used to im-
plement specialized functions,
such as translation table manipu-
lation, paging interrupt handling,
etc., that would normally be per-
formed by software. A special
Spectra 70/46 processor instruc-
tion, Function Call, is provided to
call for and execute these micro-
programmed routines.
The new 70/567 Drum Memory
Unit is currently offered for use
exclusively with the 70/46 Time
Sharing System. In this system,
pages of user programs will be
stored in on-line disk storage units,
from which they will be called to
main memory for execution. Once
in main memory; program pages
that must be temporarily dumped
to make room for higher-priority
pages are sent to the Drum Mem-
ory Unit, where they await rapid
retransmission to core when main
memory space again becomes
available. Also residing in the
Drum Memory Unit will be the
most frequently used control rou-
tines of the 70/46 Time Sharing.
Operating System (TSOS).
JULY ? 1967 Approved For Release 2005/06/03 : CIA-RbP78-03948A000100090001-8
AUERBACH
Approved or Release - - -
The purpose of so fragmenting
core storage and programs is to
permit a large number of programs
(or program segments) to reside
concurrently in core storage in a
manner that permits execution
control to be passed easily between
the resident programs, either on
an interrupt basis or a time-slicing
basis. Thus, even modest-sized
core memories can accommodate
the processing needs of a large
number of users, since only a lim-
ited number of pages of each re-
quested program require use of
core storage at any one time.
Program pages not immediately
required for execution in a Spec-
tra 70/46 system are stored in ran-
dom access secondary storage-
typically a 70/564 Disk Storage
Unit. When these program pages
are called for execution, they are
relocated by the Time Sharing Ex-
ecutive routine to any available
page in core storage. If active resi-
dent pages must be displaced by
higher-priority program pages,
they are temporarily dispatched to
the system drum unit, from which
they can be quickly called and re-
instated in any available page of
core storage.
The 512-word Translation Mem-
ory is used by the software when
assigning and relocating program'
pages to specific blocks of core
storage. The Translation Memory
sets a limit of 512 pages as the
maximum number of uniquely ad-
dressed program pages that any
program can contain. This means
that every Spectra 70/46 program
can be written , as if 2,097,152
bytes of core storage were avail-
able for its use, even though the
actual core storage unit provides
only 262,144 bytes of real storage.
The Spectra 70/46's provision of ,
about two million bytes of virtual
core storage could be expanded to
eight million bytes without sub-
stantial hardware change, but RCA
has not indicated that such an ex-
pansion is forthcoming.
Address Translation
Before executing instructions in
program pages that are dispersed
randomly through core storage, the
70/46 Processor must translate
each non-1/0 Instruction and op-
CPYRGH
erand address from its virtual ad-
dress form to the corresponding
real address that is relative to the
current location of the page in core
storage. Virtual-to-real address
translation takes place automati-
cally in the Spectra 70/46 Proc-
essor, and generally without any
increase in instruction execution
time. (Instruction execution times
for the 70/46 Processor are gen-
erally equal to those of the 70/45
Processor.) -
When the Spectra 70/46 Proc-
essor is operating in the time-shar-
ing mode, 24-bit virtual addresses
are translated to 18-bit real ad-
dresses by means of an associative
table-lookup search through the
Translation Memory portion of
read-only memory. The 12 low-
order bits (i.e., the "displacement"
field) of the virtual address are
used intact as the 12 low-order bits
of the real address. The next 11
bits of the virtual address repre-
sent the segment and page num-
ber of the program address. These
bits are translated via the Transla-
tion Memory to a 6-bit real page
address which is combined with
the untouched 12-bit displacement
field to form an 18-bit real address
that is capable of directly address-
ing up to 262,144 bytes of core
storage.
Processor Modes
Two processing modes are avail-
able for users of the 70/46 Time
Sharing System. If a specific bit is'
set in the processor's Interrupt'
Status Register and a correspond
ing bit is set within the virtual
address of an instruction or oper-
and, automatic virtual-to-real ad-
dress translation takes place and
the various microprogramed oper-
ations of the Function Call in-
struction can be utilized. This
processing mode is called the 70/
46 or time-sharing mode. All user
programs compiled by the Spectra
70/46's Time Sharing Operating
System (TSOS) software include
virtual addresses with the trans-
late bit set.
The other processing mode is
called the 70/45 mode. In this
mode all object program addresses
are treated as direct addresses,
and no address translation occurs.
Vital statistics on the new drum
include a storage capacity that
ranges between 4.1 and 16.4 mil-
lion bytes, an average access time
of 8.6 milliseconds, and a peak
data transfer rate of 333,333 bytes
per second.
70/46 Processor
The RCA Spectra 70/46 Proc-
essor is basically a 70/45 Process-
or with additional features to fa-
cilitate time-shared operations. All
features of the 70/45 Processor
are also included in the 70/46.
The principal features added to
the 70/45 in designing the 70/46
Processor include:
? Two additional banks of read-
only memory.
? Virtual-to-real address trans-
lation logic.
? An interval timer with a 100-
microsecond resolution.
? Hardware control of paging
and segmentation.
? Interrupt capabilities for in-
terval timer run-out, paging errors,
and page queuing.
? Six microprogramed operators
to control use of the translation
Memory, interval timer, and page
queuing.
? Capacity to add up to four
high-speed selector 1/0 channels
(as compared to a maximum of
three medium-speed selector chan-
nels with the 70/45).
? Memory protection that in-
cludes both Store and Fetch Pro-
tect features. The Spectra 70/46
Processor has been designed to
facilitate serving multiple users
with multiple programs. To under-
stand this design, it is necessary
to consider the concept of paging.
Paging
The Spectra 70/46 core storage
unit is divided into many blocks
of equal size called "pages." The
basic page size is 4,096 bytes, but
2,048-byte page lengths can also
be specified through the software.
Spectra 70/46 programs are com-
piled in blocks of 4,096-byte pages,
and only a limited number of
pages of each program need be in
core storage at any given instant
of program execution.
Approved For,Release 2005%06/03: CIA-RDP78-03948M1I G MAGAZINE
CPYRGH
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8 T
It will be possible to execute all
Spectra 70/45 object programs,
including the TOS and TDOS soft-
ware control routines, in a Spectra
70/46 Processor operating in the
70/45 mode, with results identical
to those obtained In a 70/45 Proc-
essor.
Instruction Set
The Spectra 70/46 Processor in-
cludes the full instruction reper-
toire of the Spectra 70/45 Proc-
essor, plus two new instructions;
Test and Set, and Function Call.
The Test and Set instruction can
be executed in either the 70/45 or
70/46 mode. This instruction is
used for bit-testing anywhere in
main memory. The Function Call
and its several microprogramed
operations (called Special Func-
tions) can be executed only in the
70/46 processing mode, and an
attempt to execute them in the
70/45 mode results in an inter-
rupt.
The purpose of the Special Func-
tions is to reduce overhead times
normally associated with time-
sharing systems by implementing
in hardware (i.e., read-only mem-
ory) functions which are usually
performed by relatively slow soft-
ware control routines. This ap-
proach to reducing the overhead
delays associated with software
operating systems has been widely
discussed but rarely used to date.
A similar use of read-only mem-
ories to implement emulators for
a
a
ties. poctra 70 corn.
r v? V V~ ^ Approved FW d1 or Releasep2005//06/03: ~ `'b 6O ( f-
JULY 1067
achieving program compatibility
with unlike computers has been
successfully accomplished by RCA
and IBM, and is currently being
developed by several other manu-
facturers.
I/O Capacity
Like the Spectra 70/45 system,
the 70/46 Time Sharing System
can concurrently execute one ma-
chine instruction, one input-output
operation on each of the installed
selector channels, and one data
transfer operation on each of the
eight subchanncis included In the
standard multiplexor channel. Up
to 256 low-speed I/O devices can
be connected to a 70/46 system
via the multiplexor channel.
Selector channels arc optional
equipment, and two, three, or four
such channels can be installed.
Each selector channel has two
trunks, permitting two I/O con-
trol units to be connected. Input/
output data is transferred into and
out of core storage in 2-byte blocks
over the Selector Channels, and
the 70/46 Processor is delayed a
maximum of 1.44 microseconds
for each block transferred. Thus,
the I/O throughput capacity of the
70/46 system is 1.4 million bytes
per second. (The 70/45 Processor,
by contrast, is delayed 1.44 micro-
seconds for each single-byte I/O
transfer, resulting in an I/O
throughput capacity that is half
that of the 70/46.) The maximum
data rate possible over a single se-
lector channel is 465,000 bytes per
second.
Time-Sharing Software
To complement the promising
hardware design of the Spectra
70/46, RCA has announced an
Impressive software package called
the Time Sharing Operating Sys-
tem. TSOS promises an efficient
random-access-oriented multipro-
graming operating system plus ca-
pabilities for providing up to 48
users at remote terminals with im-
mediate, -coriversafional access to
the central 70/46 computer. In-
cluded in the software package
will be a complete range of batch
and conversational language proc-
essors, remote syntax checking, de-
bugging and file editing systems,
an extensive data management
system, and a powerful executive
control program. RCA plans to de-
liver the entire Time Sharing Op-
erating System during the third
quarter of 1968.
The nucleus of TSOS is its net-
work of system control programs
used in the "privileged" mode by
the central Spectra 70 system. The
principal control programs in-
clude:, the resident Executive pro-
gram; the File Control Processor
(FCP) for catalog, file, and data
management; an interactive data
communications input-output sys-
tem; an interactive software de-
bugging system; and hardware
diagnostic routines.
The chief components of the
??..,FR?na C)
C)
r.a m -*sl `"' n
- - -- ..._...,.. --..... , .. ,..... .....r
" '' a
RCA's NEW SPECTRA 70/552 Alphanumeric Video D
t
THE SPECTRA 'llY/f:QA n..a., r?..e4_.d__ ^--..__ .-
A f I r n n n n, , Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Executive program will be the fol-
lowing:
? Task Scheduler, to maintain a
single task queue from multiple
input sources, including remote
terminals, and to schedule on a
priority basis the concurrent exe-
ctltion of as many programs as
the system resources will allow.
The Task Scheduler will also in-
clude a time-sharing algorithm to
control responsive time-shared op-
erations.
? Memory Manager, to control
the dynamic allocation of program
pages and to manage the use of
the more than two million bytes
of virtual storage available to the
programmer.
? Device Allocator, to manage
the pool of input-output devices
available to the system, and to as-
sign free I/O devices to the pro-
grams' device-independent I/O
requests.
? Peripheral Input-Output, to
provide optional buffering of
punched-card input files and print-
er-destined output files on random
access devices or magnetic tape
units.
? Job Control, to accept user-
specified Job Control Language
statements used to enter and de-
fine batch processing jobs at the
central 70/46 computer center.
? Command Language Control,
to enable users at remote terminals
to communicate with the central
system in either conversational or
batch n5ode. The Command Lan-
guage is an extension of the job
Control Language. It permits the
users at remote terminals to iden-
tify themselves, to specify the
tasks the system is to perform,
and to monitor the execution of
these tasks. Command Language
procedures can be prestored for
efficiency of operation.
Data Management
A major system control program
in the Time Sharing Operating
System is the File Control Process-
or (FCP). The FCP provides com-
prehensive file management and
data management systems. In the
area of file management, the FCP
maintains catalogs of all files in
the system and provides facilities
for indexed sequential access to
files stored on random access de-
vices. Options available to the user
include the capability to specify
file names and characteristics at
execution time rather than at as-
sembly or compilation time, and
the capability to share files among
users through the use of pass-
words.
Also controlled by the FCP is
the TSOS Data Management Sys-
tem. Two types of file organization
are supported: sequential and in-
dexed sequential. The Data Man-
agenicnt System provides facilities
for manipulating either type of file
by means of the following func-
tions: describe, create, access copy,
change, and delete. A file protec-
tion system prevents unauthorized
reading and writing of files in a
dynamic time-shared environment.
Processing Programs
In addition to the network con-
trol programs. TSOS provides a
large set of processing programs
available for use by the system
and problem programmers. The
TSOS processing programs include
language processors and utility or
service programs. Both the process-
ing programs and the users' pro-
grams are executed in the "non-
privileged" mode under control of
the TSOS control programs.
Language processors are sup-
plied in conversational and batch
form. The conversational language
processors include an interpretive
FORTRAN compiler that uses a
FORTRAN subset language, and
syntax checking programs for
source-level FORTRAN, COBOL, .
and assembly-language programs.
The more conventional language
processors that operate in batch
mode are a FORTRAN IV com-
piler, a COBOL compiler, a Macro
Assembler, and a Report Program
Generator.
Utility programs are also pro-
vided in both conversational and
batch modes. The two conversa-
tional service programs are the
Desk Calculator and Text Editor
programs. Using the Text Editor,
the remote user can create, modify,
and display files on a line-at-a-time
CPYRGH
(i.e., single record) basis. The
Text Editor should prove of great
value in maintaining symbolic pro-
grams and their associated data
files. The batch utility programs
include a sort/merge program, a
file maintenance routine for sys-
tem and user programs, and a
test data generator.
All software comprising the
Time Sharing Operating System is
currently scheduled for delivery
during the third quarter of 1968.
In summary, the Time Sharing
Operating System and its many
components represent RCA's an-
swer to the large-scale, random-
access-oriented software systems
offered by IBM with its Operating
System 360 and the Model 67's
Time Sharing System. TSOS sup-
plants the long-discussed Disk Op-
crating System in the RCA Spectra
70 software line. Because TSOS
includes not only time-sharing fa-
cilities, but also impressive multi-
programming and data manage-
ment facilities, it is likely that the
Spectra 70/46 Time Sharing Sys-
tem (including TSOS) will prove
attractive to some users whose
processing requirements do not in-
elude remote, time-shared process-
ing.
Compatibility
The Spectra 70/46 Time Shar-
ing System is compatible with the
RCA Spectra 70/35, 70/45, and
70/55 systems to the extent that
all instructions, character codes,
interrupt facilities, and special fea-
tures available in the nontime-
sharing systems are also available
in the 70/46 system and function
in the same manner. Therefore,
when operating in the 70/45 mode,
the 70/46 system will be able to
use the TOS and TDOS software
systems and execute object pro-
grams compiled for 70/35, 70/45,
and 70/55 systems, provided that
the programs are timing-independ-
ent and system optional features
are identical.
Functioning in the 70/46 mode
under the specialized Time Shar-
ing Operating System, the 70/46
will be able to compile all source
programs originally written to
function under TOS or TDOS in
70/35, 70/45, or 70/55 systems.
Approved For Release 2005/06/03: CIA-RDP78-03946%%0~~f ~11~ MAGAZINE
Approved or Release .
The Job Control Language (JCL)
used with the Spectra 70/46 Time
Sharing Operating System is an
expanded version of JCL as used
with the Tape and Tape/Disc Op-
erating Systems, but is not fully
compatible with the latter.
Because of the high degree of
compatibility between the 70/46
and the general-purpose Spectra
70 computer systems, users of the
70/35, 70/45, and 70/55 systems
should be able to upgrade to the
more advanced facilities of the
Time Sharing System with a mini-
mum of conversion-time difficulty.
Like the RCA Spectra 70/35,
70/45, and 70/55 systems, the
Spectra 70/46 In 70/45 mode Is
largely compatible (both in terms
of programs and data) with the
general-purpose models of the IBM
System 360 computers.
Performance
A prototype Spectra 70/46 sys-
tem is currently operational and
being used by RCA for software
development, but no detailed per-
formance estimates are available
to date. As a preliminary perform-
ance indication, RCA states that
the system will provide a maxi-
mum response time of eight sec-
onds for each of 48 interactive re-
mote terminals, assuming that all
terminals are concurrently active
and that four background pro-
grams are also being processed
concurrently,
The Spectra 70/46 hardware/
software combination as specified
is impressive and deserving of fur-
ther scrutiny by the industry as
its implementation progresses. The
70/46 Time Sharing System is not
startlingly revolutionary in its de-
sign nor extravagantly bold in its
goals (and, as such, may not ap-
peal to the industry's confirmed in-
novators). Yet, the 70/46 may
work. It appears to combine the
better features of efficient multi-
programing systems with more ad-
vanced remote data communica-
tions and time-sharing facilities.
Its approach to time-sharing is evo-
lutionary-a natural, continuous
progression from the time-tested
batch processing facilities to those
that may be better suited to cope
with the processing needs of the
future.
Table 1. Profile of the RCA Spectra 70/46 Processor
Core storage cycle time, microseconds
1.44
Bytes accessed per cycle
2
Core storage capacity, bytes
262,144
Selector channels
0, 2, 3, or 4
Multiplexor channels
1
Selector channel capacity, bytes per second
465,000
Multiplexor channel capacity, bytes per second
465,000
Maximum combined I/O data transfer rate,
bytes per second
Processor speeds, microseconds*
Fixed-point binary
1,388,888
c=a+b
25.2
c=aXb
Fixed-point decimal
81.9
c=a+b
42.2
caXb
Floating-point (short)
127.0
c=a+b
37.4
c=aXb
Floating-point (long)
67.6
c=a+b
52.6
c=aXb
211.5
,The fixed-point binary and short (looting-point operations use 32-bit binary operands. long
floating-point operations use 64-bit binary operands. The fixed-point decimal times are based
on signed 6-digit (a-byte) decimal operands.
JULY ? 1967
CPYRGH
New Hardware and Software
In addition to the 70/46 Time
Sharing System, RCA has expand-
ed and improved its product line
in other areas during the past
year. New hardware products in-
clude a controller for Burroughs,
IBM, or NCR MICR sorter-readers,
a high-speed (1,000 cps) paper
tape reader, "quictized" printers
with 96-character sets, an im-
proved CRT display device, and 7-
track magnetic tape units that
provide tape compatibility with
earlier RCA 301 and 501 systems.
More specialized hardware sys-
tems of recent development are
the Spectra 70/510 Voice Response
System, which delivers recorded
human-voice responses to on-line
inquirers at Bell System Touch-
Tone telephones, and the Spectra
70/630 Data Gathering System
(DGS), designed to gather infor-
mation at remote, point-or-transac-
tion input stations (such as badge
readers, card readers, etc.) and
transmit it to a central Spectra 70
computer system.
Software Improvements have
centered on adding random access-,
device capabilities to the Spectra
70 language processors and ex-
panding the support of data com-
munications devices. Also, the
Spectra 70 Basic Time Sharing Sys-
tem (BTSS) is being developed as
an all-software approach to limited
time-shared processing with Spec-
tra 70/45 systems.
By means of these develop-
ments, RCA has improved its posi-
tion as a worthy competitor of IBM
in all areas of the small-to-medium
scale computer market. RCA has
accomplished this feat with the aid
of a maturing market and product
planning philosophy that does not
attempt to match in every detail
the vast facilities of the IBM Sys-
tem 360, but instead attempts to
analyze the demands of the market
place and to implement whatever
facilities are required to remain
competitive, but on a scale that is
feasible and potentially profitable.
The hardware and software de-
sign of the new Spectra 70/46
Time Sharing System is an excel-
lent case in point.
Approved For Release 2005/06/03: CIA-RDP78-03948A000100090001-8
Approved.Eor Release 2005/06/03 : CIA-RDP78-0,8A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8
7 1
a8 11
a040 - H
a8o3
A103
s-4o
asa-I
dad i
:45O-I
Approved For Release. 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For W ease 2005/06/03 : CIA-RDP78-03948QP0100090001-8
TIME SHARING SYSTEM
.20,550
1052 Adapter
.232
6980 Selector Channel 1st
720
6981 Selector Channel 2nd
720
2821-1 Control Unit
1,000
3615 1100 LPM Printer Adapter
77
1990 Column Binary
103
8637 Univ Char Set Adapter
15
1403-N1 Printer
910
1416-1
Inter Changeable Train Cartridge
100
2540
Card Read Punch ,
2848-3
Display Control
435
2260-1
Display Station (8 @ 51)
408
2702-1. Transmission Control
1,075
2841
Storage Control
766
2321
Data Cell Drive
2,885
2314-1
Direct Access Storage Facility
5,410
2803-1
Tape Control
.670
2401-3
Tape Units (3 @ 810)
'2,430
39,251
Approved For Release. 2005/06/03 : CIA-RDP78-03948A000100090001-8
Approved For=Tease 2005/06/03 : CIA-RDP78-Q394& 0100090001-8
COINS
2040 - H
1.0,505
1052
Adapter
232
6980
Selector Channel
360
1052-7
65
2702-1
Transmission Control
1,075
2741-1
Communication Terminal (5 @
87.50)
437.50
2841
Storage Control
540
2311-1
Disk Storage Drive (6 @ 590)
3,540
2701-1
Data Adapter Unit
1,001
COINS System
17,755.50
TS System
39,251.00
Switching hardware
1,000.00
$ 58,006.50
17,755.50
Approved For Release 2005/06/03 : CIA-RDP78-03948A000100090001-8