RISKS DIGEST 4.44
RI...@csl.sri.com.uucp
FORUM ON RISKS TO THE PUBLIC IN COMPUTER SYSTEMS
ACM Committee on Computers and Public Policy, Peter G. Neumann, moderator
Contents:
Air Traffic Control -- More Mid-Air Collisions and Prevention (PGN)
Time warp for Honeywell CP-6 sites (P. Higgins)
GM On-Board Computers (Martin Harriman)
Loose coupling (Ephraim Vishniac)
Units RISKS and also a book to read (Lindsay F. Marshall)
Re: Unit conversion errors (Alan M. Marcum, Keith F. Lynch)
DP Ethics: The "Stanley House" Criteria (Pete McVay)
The RISKS Forum is moderated. Contributions should be relevant, sound, in good
taste, objective, coherent, concise, nonrepetitious. Diversity is welcome.
************** (EXCEPT FOR THIS HEADER!)
(Contributions to RI...@CSL.SRI.COM, Requests to RISKS-...@CSL.SRI.COM)
(Back issues Vol i Issue j available in CSL.SRI.COM:<RISKS>RISKS-i.j. MAXj:
Summary Contents Vol 1: RISKS-1.46; Vol 2: RISKS-2.57; Vol 3: RISKS-3.92.)
----------------------------------------------------------------------
Date: Thu 29 Jan 87 20:12:19-PST
From: Peter G. Neumann <Neu...@CSL.SRI.COM>
Subject: Air Traffic Control -- More Mid-Air Collisions and Prevention
To: RI...@CSL.SRI.COM
There were several collisions recently that are worthy of note here.
A twin-engine 18-seat Metroliner and a single-engine private plane
collided near Salt Lake City on 15 January 1987. All 10 aboard killed.
The small plane had no altitude transponder.
An Army twin-enginer turboprop collided with a twin-engine business plane
near Independence, MO. All 6 people killed. 19 January 1987. Both
planes had altitude transponders, but controllers said they did not see
the altitude data on their screens.
A six-seat regional airliner and a single-engine private plane grazed
each other near Westerly, RI. No one hurt. 19 January 1987.
The FAA is contemplating extending its forthcoming Mode-C regulations to
include commuter planes as well as mainliners. [Source: SFChron 25 Jan 87]
------------------------------
Date: Thu, 29-JAN-1987 10:30 PST
From: <PHig...@UCIVMSA.BITNET>
To: <RI...@CSL.SRI.COM>
Subject: Time warp for Honeywell CP-6 sites
All Honeywell CP-6 sites running version C01 of CP-6 suddenly entered a time
warp Wednesday morning. The Front End Processor (FEP) suddenly thought it was
December of 1968, but the host still knew the correct time and date. It
turns out that the sign bit of the word containing the time finally got set.
Unfortunately, that word appears to have been declared as a signed number
rather than an unsigned one. (How you could ever have a negative time is
beyond me.) Since the base time for CP-6 is January 1, 1978, we suddenly
scooted back in time to 1968. The problem was first reported by a CP-6
site in Germany at 5:23am Pacific Standard Time.
What impact did this have on CP-6 users? Those using programs running
solely on the host weren't affected, though the login message gave the
wrong time and date. Those using the Transaction Processing (TP) features
of the FEP, however, discovered that incorrect dates were entered into
their databases on the host. CP-6 sites are now manually correcting the
bad data.
The problem was fixed by early Wednesday afternoon and a patch was made
available to Honeywell's customers.
By the way, one CP-6 site determined that the time stamp will overflow
at 15:26:07.35 on October 11, 1999. Mark that down on your calendars!
Here's the message Honeywell sent to its customers. (A STAR is a problem
report.)
Sent: 01/28/87 06:39 Rcvd: 01/28/87 06:53 Number: 68
To: CUSTOMERS,CP-6 FOLKS From: (deleted)
Subject: FEP timestamp problem
Yes, good morning y'all. The CP-6 interpretation of the level-6 timestamp
seems to have started to pickup a sign bit today, so every site in the
world, regardless of patch revision or system revision is happy, happy,
happy. Star 32173 has been generated to track this problem at severity
one. If you want to be on the list of people to be notified as soon as a
fix is available, please build a note on that star. Sorry 'bout that.
------------------------------
Date: Wed, 28 Jan 87 15:24 PDT
From: Martin Harriman <"SRUCAD::MARTIN%sc.intel.com"@RELAY.CS.NET>
To: gzt...@mc.lcs.mit.edu, ri...@CSL.SRI.COM
Subject: GM On-Board Computers
The reason that the ROMs on the GM Engine Control Module do not come with
the replacement unit is that they are specific to the automobile--they vary,
depending on the particular automobile model, engine, and transmission.
This is no different from a number of other components (distributor innards
in older cars, for instance). The ROMs are available (on special order)
as replacement parts; you need to know the exact configuration of your
GM car to order them (that is, the VIN, the engine code, and the transmission
code--I don't think you need the complete set of option order codes, though
these are usually available on a sticker buried in the trunk). The ECMs are
generic (there have only been a few major revisions of the basic module), and
the ROMs are (extremely) specific--so it is not possible to stock the complete
set of ROMs, nor to stock ECMs with ROMs installed.
There are several specialized tools available for ECM based diagnostics.
For field use, several companies make special tools which plug in to a
connector under the dash, and communicate with the ECM to monitor engine
parameters and diagnose faults. This is an amazingly powerful tool for
diagnosing engine problems; you can, for instance, see if the engine tends
to run rich or lean in one particular regime by reading out the current
"block learn mode" matrix from the ECM (this is a set of fudge factors the
ECM keeps so it can guess at the correct fuel delivery for your particular
car and engine). All GM dealers should have such tools, and (presumably)
know how to use them.
Incidentally, you can read out some of the ECM diagnostics with nothing
fancier than a bent paperclip; the GM shop manuals give all the details.
This is most useful in a case where the ECM has already found a problem,
and illuminated the "Service Engine Soon" light (the ECM checks all its
sensor values for "reasonableness"; if things don't seem right, it
complains).
------------------------------
Date: Wed, 28 Jan 87 20:25:34 est
From: <decvax!wanginst!wang!eph...@ucbvax.Berkeley.EDU>
To: CSL.SRI.COM!RISKS
Subject: Loose coupling
In Risks 4.42, Alan Wexelblat asks about the applicability of the
principle of "loose coupling" to computer systems. I think the
principle is a valuable one. Herewith, a brief study in contrast.
My present employer, Wang Labs, makes a variety of computer systems. The
Wang VS series are conventional minicomputers. That is, they have a cpu
which runs user tasks, with a conventional OS. The Wang OIS is a
loosely-coupled system in which a central file server (the OIS "master")
supports a collection of workstations and peripheral devices. The VS, which
is probably no better or worse than most computers of its class, suffers
occasionally from task crashes and OS crashes. Installation of new
peripherals or major new software generally requires an IPL or two. On the
OIS, all user code runs in the workstation. If your workstation (or other
peripheral) crashes, the most that's required is to cycle power on the
device. The master sees the power-up, and reloads you. *All* OIS software
except the master code can be re-installed without an IPL. Peripherals can
be installed simply by plugging them in.
In a development environment, VS's are sometimes reloaded hourly in
order to change software, change configuration, or recover from crashes.
(Released software, of course, is orders of magnitude more stable.)
OIS's? The last time mine was IPLed was to recover from a mechanical
disk failure, months ago. Master crashes are practically unheard of.
Generally, I think loose coupling presents an invaluable opportunity
for bullet-proofing of components. It becomes possible to validate
your input and to recover from external problems, only when "input" and
"external" are well-defined terms. Let the lines be drawn.
Disclaimer: These are my own opinions about Wang products. Other
Wang employees, salesmen, and customers have their own opinions.
Ephraim Vishniac, decvax!wanginst!wang!ephraim
------------------------------
From: "Lindsay F. Marshall" <lindsay%cheviot.new...@Cs.Ucl.AC.UK>
Date: Tue, 27 Jan 87 16:29:56 GMT
To: ri...@csl.sri.com
Subject: Units RISKS and also a book to read
Another classic case of mistaken magnitude is documented in a variety of
books on the CIA. When they were carrying out their infamous LSD experiments
they heard that Sandoz had for sale 22lbs of LSD and being so afraid that
the Russians would buy, put $250,000 in a case and went shopping. The people
at Sandoz looked very puzzled - they had only ever made about the 0.5 oz of
the drug. Someone in the CIA Swiss office didn't know the difference between
milligrams and kilograms.
I just finished a quite entertaining book featuring a computer crimes
investigator of the year 2000. The technical stuff is OK (for a change) and
the basic idea behind the plot is quite feasible (and scary!!) from a RISKS
point of view. The book is :
Downtime
by Peter Fox
ISBN 0-340-39362-9
Published 1986 by Hodder & SToughton (in the UK at least)
Lindsay
------------------------------
Date: Tue, 27 Jan 87 10:45:25 PST
From: mar...@Sun.COM (Alan M. Marcum, Consulting)
Subject: Re: Unit conversion errors
To: ri...@csl.Sri.COM
A unit conversion error (pounds and kilograms, if I recall) was a major
contributing factor in the Air Canada 767 flameout incident a few years
ago. The jet ran out of fuel during cruise; the pilot also flew
sailplanes, the co-pilot trained near (90 miles away from) the flameout
site, and they were able to land safely.
Alan M. Marcum Sun Microsystems, Technical Consulting
mar...@nescorna.Sun.COM Mountain View, California
------------------------------
Date: Wed, 28 Jan 87 00:04:03 EST
From: "Keith F. Lynch" <KFL%MX.LCS....@MC.LCS.MIT.EDU>
Subject: Units
To: brian%kelpie.newc...@CS.UCL.AC.UK
cc: KFL%MX.LCS....@MC.LCS.MIT.EDU, RI...@CSL.SRI.COM
I think it goes like this:
Power American British Metric
of ten name name prefix
3 Thousand Thousand Kilo
6 Million Million Mega
9 Billion Milliard Giga
12 Trillion Billion Tera
15 Quadrillion
18 Quintillion Trillion
21 Sextillion
24 Septillion Quadrillion
27 Octillion
30 Nonillion Quintillion
The problem is not that different names are used, but that the same
names are used for very different numbers. This is why metric
prefixes have caught on. For instance a thousand million electron
volts is now called a GEV (for Giga) rather than a BEV (for Billion).
Unfortunately, the metric prefixes don't go very far. In any case,
they are hard to remember. And they are no longer always unambiguous,
at least in the computer world, where the prefix Mega may mean 1,000,000
or 1,048,576 or even 1,024,000.
The best solution is to use the exponents of ten. Instead of GEV,
just say 1E9 EV. This is catching on rapidly, perhaps due to computers,
which are more easily programmed to say 1.02E+09 than 1.02 GEV, etc.
You are right, Milliard is not used in the United States. Actually, the
highest name I ever hear is trillion. People who speak of quadrillions tend
to get funny looks.
[Wait until our national budget gets there in a few years! PGN]
Science meets engineering where I work, too. This results in code where a
distance is always stored as millimeters, but is output and read in in mills
(thousandths of an inch). Similarly with degrees C and F. Sometimes fully
a third of my programming time is taken up with making sure incompatible
units don't mix, and making extra sure whether I multiply or divide by a
conversion factor or its inverse, etc.
I discovered a bug in someone else's code where a unit was BTUs per cubic
foot per degree F and was later used as calories per cubic centimeter per
degree C. I spent a whole day fixing this, before working out the actual
conversion factor to put in the constants section - and the conversion
factor was exactly 1.
...Keith
------------------------------
Date: Wednesday, 28 Jan 1987 06:48:37-PST
From: mcvay%telco...@decwrl.DEC.COM (Pete McVay -- VRO 5-1/D7 --DTN 273-3106)
To: ri...@csl.sri.com, mcvay%telco...@decwrl.DEC.COM
Subject: DP Ethics: The "Stanley House" Criteria
In 1976, the Canadian government sponsored a meeting in Quebec at the
"Stanley House", composed of top data processing experts and philosophers.
The meeting specifically addressed the issue of ethical conduct in the
computer industry. The list that follows was extracted from an article
published that year in SCIENCE Magazine (the official magazine of AAAS).
Unfortunately, I do not have the date of the magazine.
I have not heard of any followup on these criteria, either in a
discussion on computer risks or ethics, or as any meaningful attempt
to implement them. I present them here to RISKS DIGEST for comments
by other readers--and perhaps someone has some later information?
Stanley House Criteria for
Humanizing Information Systems
1. Procedures for dealing with users.
A. The language of a system should be easy to understand.
B. Transactions with a system should be courteous.
C. A system should be quick to react.
D. A system should respond quickly to users (if it is unable
to resolve its intended procedure).
E. A system should relieve the users of unnecessary chores.
F. A system should provide for human information interface.
G. A system should include provisions for corrections.
H. Management should be held responsible for mismanagement.
2. Procedures for dealing with exceptions.
A. A system should recognize as much as possible that it
deals with different classes of individuals.
B. A system should recognize that special conditions might
occur that could require special actions by it.
C. A system must allow for alternatives in input and
processing.
D. A system should give individuals choices on how to deal
with it.
E. A procedure must exist to override the system.
3. Action of the system with respect to information.
A. There should be provisions to permit individuals to
inspect information about themselves.
B. There should be provisions to correct errors.
C. There should be provisions for evaluating information
stored in the system.
D. There should be provisions for individuals to add
information that they consider important.
E. It should be made known in general what information is
stored in systems and what use will be made of that
information.
4. The problem of privacy.
A. In the design of a system all procedures should be
evaluated with respect to both privacy and humanizing
requirements.
B. The decision to merge information from different files
and systems should never occur automatically. Whenever
information from one file is made available to another
file, it should be examined first for its implications
for privacy and humanization.
5. Guidelines for ethical system design.
A. A system should not trick or deceive.
B. A system should assist participants and users and not
manipulate them.
C. A system should not eliminate opportunities for
employment without a careful examination of consequences
to other available jobs.
D. System designers should not participate in the creation
or maintenance of secret data banks.
E. A system should treat with consideration all individuals
who come in contact with it.
[This seemed worth drawing to your attention, although it might also
be suited to Human-Nets and Soft-Eng. But to prevent subsequent
discussion from wandering all over the place, let's see if we can
constrain it to RISKS-related matters. Thanks. PGN]
------------------------------
End of RISKS-FORUM Digest
************************
-------