PROJECT ORACLE
Document Type:
Collection:
Document Number (FOIA) /ESDN (CREST):
CIA-RDP80-01794R000100230024-0
Release Decision:
RIPPUB
Original Classification:
K
Document Page Count:
21
Document Creation Date:
December 16, 2016
Document Release Date:
October 16, 2004
Sequence Number:
24
Case Number:
Publication Date:
November 11, 1975
Content Type:
MFR
File:
Attachment | Size |
---|---|
CIA-RDP80-01794R000100230024-0.pdf | 1.01 MB |
Body:
Approved Release 2004/10/28 :-CIA-RDP80-014R000100230024-0-
11 November 1975
MEMORANDUM FOR THE RECORD
SUBJECT Project ORACLE
On 29 October 1975 of the
Agency me with the AMP representatives, Mrs. Lemos and
.Messrs. Wood and Slingland for the purpose of determining the
direction and status of AMPEX's design of the ORACLE Mass
Storage. System. There were four areas discussed in detail
that we felt were not being properly designed.. We asked
AMPEX to tell us clearly what they were doing. The remainder
of this paper is devoted to our understanding of the designs,
there implications, commentaries and references that explain
our concern, and a statement of our requirements for each
design area.
I. Storage Control Processors accessing Transport Drivers
Hardware Design
Mr. Wood stated that the current AMPEX desgn permitp
one an ne ,tq ontro Prace~sor (SCP) to be enabled
at an iven time to access anr.of Sher Transport Drive . j Q.
TNe ' CP's aTTD'sare `cabled together in such a way that when
one SCP is in control of the TD's, the other SCP is blocked
from all access to any of the TD's by a hardware interlock.
In order for the second SCP to gain access to the TD's, the
first SCP must relinquish its control and then it is possible
for the second one to take over. Once the second SCP is in
control the first one is inhibited from all access'to the TD's.
Any interrupt issued by a,TD is serviced by whichever SCP is
currently in control. The TD does not know which SCP should
handle the interrupt, it is expected in this design that this
type of problem can be adequately solved at the SCP level.
The drawing below illustrates how these hardware elements are
tied together logical..
STAT
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved F Release 2004/10/28 CIA-RDP80-014R000100 30024-0
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
TD 6
Approved ,pr Release 2004/10/28: CIA-RDP80-01%94R000100230024-0
Implications
The above design gives control, at any given time, of
all TD's to a single SCP. If dual SCP control of TD's is
desired, the two SCP's could flip-flop by means of software
logic. A flip-flop technique would permit segregating or
sharing those functions in the two SCP's that require an
SCP-TD association. Implementation of such a scheme adds
complexity to the software logic which would have to reside
in both SCP's. The areas particularly affected are hand-
shaking protocols, timing, and recovery. It also requires
that a wise choice be made initially as to which functions are
shared or segregated between the-SCP's. The choice is
critical since once designed and implemented, subsequent
changes would not be easy.
AMPEX has chosen not to develop a flip-flop design. They
have ruled out any type of dual SCP operation, the second SCP
is for back-up only. The hardware design does not permit
selective accessing of TD's by both SCP's concurrently and
the software design requires that-all accesses of TD's be
made from only one SCP.
There are advantages to the design chosen by AMPEX. A
single SCP controlling all the TD's needs only the simplest
software and somewhat reduces maintenance or the hardware.
There is a major disadvantage to the design. It is
impossible to use the redundant equipment for software
development and testing. Although the backup SCP and TD's
can be put off-line there is no way of using them independent
of the production system. This means that once the system is
in production, fixing of existing bugs can be done only by
taking down the production system. Assuming that the production
system cannot be down very often such work would require long
periods of elapsed time.
Larger efforts such as minor improvements to the existing
system, development of the new functions, e.g. time sharing,
and adjusting interfaces to new levels of operating systems
become impossible unless more hardware is added or all of this
work is contracted out.
Approved For Release 2004/10/28: CIA-RDP80-01794R000100230024-0
Approved Fire Release 2004/10/28: CIA-RDP80-01 *AR000'100230024-0-
It should be obvious to the reader that all of the above
mentioned tasks are necessary. The delivered system will not
survive unless it is enhanced and modified'to accommodate new
operating systems.
The AMPEX Design versus Contract Specifications
The AMPEX design of the accessing paths among the SCP's
and TD's does not comply with the specifications stated in the
AMPEX Proposal TBMP 73-1 and in the Mass Storage System Design.
The following paragraphs trace the reasons for our conclusion.
Only documentation referenced by the contract is used as a
basis for argument. We have had-many conversations with _MPEX
during preparation of the specification that leave no doubt, -' '--:; _,..
however, it is felt they cannot be used as part of the record.
Reference: AMPEX Proposal, Page 2-1, Paragraphs 3 and 4.
The TBM* Memory System is highly modular in construction.
System capabilities can be varied-over a wide range by
configuring the system to include different numbers of each
of five basic building blocks: Transport Modules (storage
capacity), Transport Driver Modules (multiple seek/search),
Data Channel Modules (internal data thruput), Storage Control
Processor (file management) and External Data Channel Processor
(data interface handling). Stora e capacity is available over
a range from 1011 bits to 3 x 101 bits in 1011 bit increments,
while sustained thruput can be specified up to 4.2M bytes per
second in .7M bytes per second increments.
Switching matrices interconnect the hardware modules.
They are constructed to allow multi-path access to any of the
hardware modules. Systems configured with redundancy for all
of the five basic building blocks therefore offer highly
desirable degradation characteristics since no single unit
failure brings the entire system down.
Commentary:
The above paragraphs introduce. the general philosphy
of the system. Namely it is intended to be modular,
expandible, highly interconnected, and that given redundancy
of equipment the system will be operable during the failure
of one of the basic units. The statement concerning single
unit failure-is important. AMPEX has used a single line to
electrically interconnect the SCP's and TD=s. Failure of this
line will bring down the entire system. No provision for
redundancy has been made by AMPEX. Thus the design is incon-
sistent with the specification.
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved Rev Release 2004/10/28 CIA-RDP80-01 R000100230024-0
Reference: AMPEX Proposal, Page 2-41, Paragraph 3.
Availability 24 Hours per. Day
'Scheduled maintenance of the MSS is performed on a
module basis, and there is no requirement for scheduled down-
time of the complete system. Preventive maintenance is usually
conducted during off-hours. The hardware utilization does
not exceed 12% for any module during the night shift (1800-
0800). The Transport utilization is less than.2% during this
period leaving more than ample for maintenance. Preventive or
corrective maintenance can be conducted in the off-line mode
concurrent with on-line operations. See Section 13.1 for
maintenance procedures. -
Commentary: ..
On-line refers to the array of hardware devices that
are-in use for production operations. Off-line refers to
those hardware devices that are logically and sometimes
physically disconnected from the on-line system. The paragraph
above calls for maintenance of off-line devices concurrent with
on-line operations. Some maintenance and hardware tests
require that an SCP access a TD. When an SCP and a TD are
both put off-line, the AMPEX design will not permit. the needed
access. Thus the specification that requires off-line maintenance
concurrent with on-line production cannot be satisfied because
of the way AMPEX has cabled the SCP's and TD's.
Reference: Mass Storage System Design, page-19, paragraph 1
TBMTAPE'initialization is performed by a stand-alone
SCP and TDP. Initializing a TBMTAPE begins with the
ecording`and testing of three longitudinal tracks: the
Address, Tally, and Control Tracks. This is followed by
search testing to determine tape packing characteristics.
Finally the wearing qualities of the tape are tested by
repeated reads of a single block.
Commentary:
The key point is in the first sentence which pairs a
stand-alone SCP with a TDP (Transport Driver Processor).
Approved For Release 2004/10/28: CIA-RDP80-01-7948000100230024=0- -
Approved Nor Release 2004/10/28 : CIA-RDP80-01` R00010g230024-0
The TDP is an integral part of the TD. The term stand-alone
is defined in the same document as an SCP be.ng off-line to
the Mass .Storage System. The tape initialization process
requires that a stand-alone SCP access a TD. This cannot
be done given the described AMPEX design because the on-line
SCP would have control of all of.the TD's. the on-line SCP
cannot transfer control of the TD's to the other SCP when it
is in a stand-alone condition. Thus the referenced specifi-
cation cannot be satisfied-because the stand-alone SCP is
unable to access a TD.
Agency Requirements concerning SCP access to TD's
1. The Storage Control Processors (SC''s) must be able
to access the Transport Drivers (TD's).in such a manner that
given redundancy of SCP's and TD's, no single failure will
cause the entire Mass Storage System to be inoperable.
2. A stand-alone SCP must be able to access a TD so
that a TBMTAPE can be initialized.
3. An SCP must have access to the TD'a such that
off-line maintenance of an SCP can be concur_ent with on-line
operation of the Mass Storage System.
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved.h Release 2004/10/28 CIA-RDP80-01 4R00010230024-0-
230024-0-
II. Usage of Two Storage Control Processors in the.Mass
Storage System
'Software Design
Mrs. Lemos stated that AMPEX is designing and developing'
the Mass Storage System (MSS) such that only one Storage
Control Processor (SCP) will be active. The second SCP's
role is purely back-up and will be switched into the system
when a failure occurs in the first SCP. .
Implications
The above design greatly simplifies the software logic
needed for the Mass Storage System. If a single SCP can
drive the system such that system thrrughput can be maintained
as specified then we cannot say that two active SOP'S is
superior to a single SCP system
---It is difficult to reconstruct why a dual-SCP system was----
originally specified. The major problem here is that the
revision of the design was done unilaterally by AMPEX. There
were no prior joint discussions on this matter, we were simply
informed of the AMPEX decision.
The AMPEX Design versus Contract Specifications
The AMPEX design of using a single active SCP for the
MSS rather than two active.SCP's does not comply with the
specifications stated in the AMPEX Proprosal TBMP 73-1 'and
in the Mass Storage System Design. The following, paragraphs
trace the reasons.
Reference: AMPEX Proposal, Page 5-1, Paragraphs 1 and 2.
Control of MSS is divided into three parts and is
per by three sets of computers. Overall system control
is provided by the Storage Control Processor Complex consisting
of one or more SCP's. The TBM* Memory System configured for
the ORACLE application comprises two identical SCP's.
The SCP Complex communicates with subscribing host CPU's,
performs file management and space allocation functions,
defines the necessary functions and transmits the corresponding
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved For Release 2004/10/28 : CIA-RDP80-0144R00010023Q024-0
commands to the other controllers in the TBM* Memory System.
During normal operations, one of the SCP's acts as the
Master (SCPM) exercising overall TBM* Memory System control
while the second one operates in a Slave mode (SCPS) per-
forming file management and space allocation functions.
Commentary:
The specification calls for two concurrently active
SCP's having different but complementary functions.
Reference: AMPEX Proposal, Pages'5-2, Paragraphs 2 and 3.
The SCPM exercises overall system control. It allocates
tasks to the other processors within the MSS, keeps an audit
trail and a file management trail for all tasks entering
the MSS, keeps an activity log for each file and for hardware
resources, monitors overall MSS operation, and automatically
switches to a degraded mode if hardware resources become
unavailable.
The SCPS normally performs space allocation for the DSS,?and
maintains the?TBMCATALOG consisting of the Master File
Directory of all files stored on TBMTAPE and the On-Line
File Directory for all mounted TBMTAPE's in the DSS. The
SCPS thus performs most of the file management functions
under the supervision of SCPM.
Commentary:
These paragraphs give further detail about the functions
to be allocated to each SCP.
References: .AMPEX Proposal, Page 5-2, Paragraph 4.
Page 5-11, Paragraph 5.
Page 5-12, Paragraph 2.
Page 5-14, Paragraph 2.
Page 2-10, Paragraph 5, 6, and 7.
Commentary:
These paragraphs are not reproduced here. All specifi-
cally discuss the concurrent use of a master and slave SCP.
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved Vir Release 2004/10/28: CIA-RDP80-01'lt4R00010d~230024-0G
Reference: Mass Storage System Design, Page 113, Paragraph 2.
The'MSS software provides the capability to attach three
consoles to each SCP in the system. (See Section 3 for the
hardware configuration). During MSS operation, one SCP is
considered to be the master SCP; the other the slave. The
consoles attached to the master SCP are used to issue commands
to the system. Other consoles are attached to various sub-
components as required for maintenance. Hosts connected to
the master SCP can also issue certain commands.
Commentary:
This paragraph gives detail about activity on the master
Summary
It should be clear that the AMPEX design which eliminates
the master-slave SCP feature is contrary to the specifications
referenced by the contract.
Agency Requirements Concerning Dual SCP's.
Technically, it is not clear that the Mass Storage System
requires more than a single active SCP. We must verify that
the single mode will not create excessive averhead, however,
before permitting AMPEX to-continue.
Other consideratiors concerning AMPEX averal7, performance
should preclude simply permitting the specification change.
We should not forget that the single or dual decision provides
us with good'leverage that can be used to advantage elsewhere.
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
Approved FooRelease 2004/10/28: CIA-RDP80-01711 8000100230024-0
.III. Access of Data Private to the Mass Storage System
Software Design
The Mass Storage System (MSS) keeps extensive information
about the identification, location, and status of the data
files under its control. This information along with other
internal MSS records is stored on two disks that are private
to the MSS. Only the MSS can directly access and use the
information.
The Storage Control Processors (SCP's) and the External
Data Channel Processors (EDCP's) are the MSS hardware modules
that access these private disks.. Presently, AMPEX has
designed the software such that the SCP's require a dedicated
disk control unit to access the private disks. Dedicated
here, means that only the SCP's can use the control unit.
When the EDCP's require access.it must be gained via another
control unit.
The figure below shows the AMPEX design.
SCPl
DISK
CONTROL
UNIT
SCP2
EDCP1
DISK
CONTROL
UNIT
EDCP2
RIVATE
DISK1
RIVATE
DISK2
Approved For Release 2004/10/28 : ?CIA-RDP80-01794R000100230024-0
Approved R&WRelease 2004/10/28: CIA-RDP80-0118000100230024-0
Implications
The *use of a dedicated controller by the SCP's requires
that a second one be made' available in the event the first
one fails. If this were not done the failure would result
in the MSS being inoperable. Thus the impact of the AMPEX
design is limited to the additional funds required for a
spare controller and the space needed for its placement.
The AMPEX Design versus Contract Specifications'
The AMPEX design which requires a dedicated disk control
unit to service the SCP's is contrary to what has been
specified in the AMPEX Proposal TEMP 73-1 and in the Mass
Storage System Design. The MSS Design calls for two disk
control units with dual access features to he used to serve
any of the SCP's and EDCP's. The AMPEX Proposal specifies
a single control unit with a Four Channel Switch to serve
the -SCP's and EDCP's. Specific references and drawings are
shown below. - - -- -
Reference: AMPEX Proposal, Page 2-24, Table entitled,
"1012 Bit TBM* Memory System Hardware Configuration"
CGS
2 SCP's (Storage Control Processors)
2 EDCP's (External Data Control Processors)
1 3330 Control Unit with 2 Spindles and a Four
Channel Switch
6 Host CPU message interface links co each SCP
8 Data interface links on each EDCP,, providing
for 16 shared device controllers vich as 3830's,
3803's, channel to channel adapter, etc. in any
combination.
Commentary:
This table lists the hardware needed fc-r a 1012 bit system.
The item - "1 3330 Control Unit with 2 Spindles and a Four
Channel Switch" - states very clearly that cniy a single
disk control unit is needed.
Approved For Release 2004/10/28 :.CIA-RDP80-01794R000100230024-0
Approved F Release 2004/10/28 :CIA-RDP80-01 04R000100230024-0
Reference: AMPEX Proposal, Page 2-36, Paragraph 2
The 3330 disk system. internal to CCS can be accessed
from either of the SCP's and EDCP's. One spindle will nor-
mally be dedicated to the MSS Catalog and not accessible
by the EDCP's. The second spindle serves as backup, and
can also be used for internal data staging, diagnostics
and maintenance functions,' etc.
Commentary:
The first sentence refers to the private disks for use
by the MSS. It provides further-elaboration of the table
described in the previous Reference. -
References: AMPEX Proposal, Page 2-25, Table entitled,
"10" Bit Initial TBM* Memory System Hardware
Configuration"
AMPEX Proposal,- Page 3--1--Table entitled;
"Hardware Required for the Initial MSS'
AMPEX Proposal, Page 4-1, Table entitled,
"Hardware Required for the Complete MSS"
Commentary:
The above references are not reproduced here, but all
of them list a single disk control unit to be used to service
the SCP's and EDCP's.
Reference: Mass Storage System Design, Page 9, Figure 3.
Approved For Release 2004/10/28 :. CIA-RDP80-01794R000100230024-0
3 ,?30Z
Approved For Release 20041 0/28: CIA-RDP80t01794R000100230024-0
JAI rA I
F
;7.5' 5
33301
33 301
Figure 3. ORACLE BSS Configuration, Oct - Dec, 1975
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0
3 $'302
S C .'O.
Z-
Approved F"Release 2004/10/28: CIA-RDP80-0171000100230024-0
.Commentary :
The above figure shows two disk control units to provide
all EDCP's and SCP's with access to the private disks. There
is no control unit allocated specifically for the'SCP's.
Summary :
All of the above references prove that. the AMPEX design
.does not comply with the specifications.
Access to the private disks of the MSS must be provided
by disk control units that are shared among the SCP's and
EDCP's as specified in the Mass Storage System Design (Speci-
fication) dated March 19, 1'975.
Agency Requirement
Approved For Release 2004/10/28 : CIA-RDP80-01794R000100230024-0