The Risks Digest

The RISKS Digest

Forum on Risks to the Public in Computers and Related Systems

ACM Committee on Computers and Public Policy, Peter G. Neumann, moderator

Volume 11 Issue 73

Tuesday 28 May 1991

Contents

o Viper
Brian Randell
o Maintenance of constants
Douglas W. Jones
o The RISKS of Posting to the Net
Mark Thorson = mmm
o Re: Risks of posting on the NET
Jim McLeod
Ellen Spertus
Mike Olson
o Re: Replicated Errors
Robert McClenon
o Re: Are fuzzy controls risky?
Rob Horn
o AT&T billing problem: "computer error"
Charles P Pfleeger
o Caller ID in commercial applications
Walter Roberson
o Info on RISKS (comp.risks)

Viper

<Brian.Randell@newcastle.ac.uk>
Tue, 28 May 91 17:44:52 BST
From the UK national newspaper, The Independent, 28 May 1991 - reprinted in its
entirety.   Brian Randell

MOD IN ROW WITH FIRM OVER CHIP DEVELOPMENT

A British company that hoped to market one of the world's most sophisticated
computer chips is to be wound up next week amid acrimonious allegations about
the Ministry of Defence's role in commercialising a technology developed at
taxpayers' expense.

The episode highlights the continuing failure of the Ministry of Defence to
foster civilian spin-offs for the products of its military research and
development programme, which last year cost the taxpayer some (pounds) 2.2bn.
Throughout most of the past decade, the Government has spent more money on
defence than on civil research and development.

The chip, known as Viper, was designed by scientists at the Royal Signals
Research Establishment in Malvern. It is the most advanced chip, designed for
use in "safety critical" applications - such as nuclear reactor shutdown
systems, driverless trains or aircraft controls - where lives depend upon
faultless operation.

When the Worcester-based company Charter Technologies goes into voluntary
liquidation on 4 June, no British company will be left able to provide
potential customers with software to program the Viper chip or provide back-up
support for its use. The company issued a writ against the Ministry of Defence
this year for alleged negligent misrepresentation of the chip's capabilities
and of its potential market.

The ministry denies the company's allegations and lodged a defence. The MoD had
required the company to post security of (pounds) 75,000 to cover the
ministry's legal costs should Charter Technologies have lost its legal action.
The company ceased trading on 2 May 1991 and is no longer pursuing its legal
action.

Digby Dyke, managing director of Charter Technologies, said yesterday that the
company was forced to seek voluntary liquidation after the MoD declined to
extend or renew other contracts, unconnected with the dispute over the chip,
and because of the losses incurred over the Viper project.

When the chip was developed in the late 1980's, the then director of the Royal
Signals Research Establishment, Nigel Hughes, described it as "the first
commercially available microprocessor with a proven correct design".  Modern
microprocessor chips contain such complex circuitry that it is often not
possible to demonstrate that the design is completely free of error. As a
result, microprocessor developers are increasingly turning to the use of formal
mathematics to verify that designs are free of errors.

In 1987, the MoD granted a licence to Charter Technologies to develop software
to exploit the chip's capabilities. But this January, the company issued a writ
alleging that the chip's design had not been proven, and as a result its money,
manpower and time were wasted. The company was alleging, in effect, that the
mathematics were not exhaustive.

Although Viper was developed by the MoD and released for commercialisation four
years ago, and although a new defence procurement standard for safety-critical
equipment, known as 00-55, appears to favour mathematically proven designs,
Kenneth Carlisle, the Under-Secretary of Defence Procurement, told the House of
Commons last week that "Viper is not currently used in any safety-critical
computer systems controlled by the MoD".

The only civilian customer for the technology has been the Australian National
Railway Commission, which at the end of 1988 adopted Viper as the core of a new
signalling system. The Australians have now gone so far with developing the
system that they would have difficulty switching to another technology.

Computing Laboratory, The University, Newcastle upon Tyne, NE1 7RU, UK
Brian.Randell@newcastle.ac.uk   PHONE = +44 91 222 7923 FAX = +44 91 222 8232


Maintenance of constants

Douglas W. Jones <jones@cs.uiowa.edu>
Tue, 28 May 91 12:18:48 CDT
In RISKS-08.19, I described a situation where, due to the cost of modifying
a contract, it was less expensive to maintain dead code than to eliminate
the dead code.  As a result of this story, I have been approached with a
similar story from AT&T; the employee who passed on the story asked that I
not mention anyone's name.  Here is the story:

It seems that the current generation of electronic switching systems coming out
of AT&T are controlled by upwards of a million lines of C code.  One of the
results of this is the need to maintain the constants in the program.

You might think that constants are constants, so this should be simple, but
this is not so.  There are two problems that make this a headache big enough to
require significant manpower.

Problem 1) Not all C constants are inherently constant.  Consider string
constants.  These are just arrays of characters with appropriate initial
values.  Once such an array is passed to a procedure, the fact that it is
constant is no longer known.  Of course, on some computers, memory protection
mechanisms could be used to make all strings read-only, but this is outside the
scope of the C language and would introduce the possibility of errors in
constant usage stopping a system that is not allowed to stop.  Thus,
programmers must be dedicated to auditing every use of every constant that is
stored in memory to assure that it is used as a constant.

Problem 2) Because C has a very weak type system, named numeric constants are
used instead of enumerated types.  As a result, it is quite possible to use the
wrong constant and never detect it because the constant has the right value.
For example, if one enumeration is (Red, Green, Blue), and another is (Apple,
Bananna, Cantalope), your program may run correctly with MyColor = Cantalope,
but later, when you add BlueBerry to the second enumeration, the value of
Cantalope (which was 2) has changed to 3, and your program will stop working
correctly.

Pascal and Ada are both able to enforce the constancy of constants at compile
time, and they go a long way towards eliminating problems with constants that
coincidentally have the same value.  (The latter problem cannot be completely
eliminated in any language.)

As I understand the story, there is an actual team of programmers at AT&T who
are responsible for auditing constant usage, and nothing else.
                                                 Doug Jones


The RISKS of Posting to the Net (RISKS-11.71, -11.72)

<mmm@cup.portal.com>
Tue, 28 May 91 16:56:40 PDT
Brinton Cooper <abc@BRL.MIL> wonders how I got access to the information on the
destruct system of the Poseidon and Polaris A3.  It was in a book that was
published by Lockheed as a training manual.  The book had no classification
markings, in fact the one place in the book where it referred to classified
information was a pointer to a separate, classified document which I did not
have.  I found it at a sale of used books.  It is not too uncommon to find
classified material in used bookstores, though I repeat this was _not_
classified.

Phil Agre <phila@cogs.sussex.ac.uk> asks whether I had any moral qualms about
talking to the FBI.  Judging from his letter and most of the e-mail I've
received, it's apparent that many people consider the FBI to be the enemy of
the Constitution and the Bill of Rights, rather than one of its defenders.
Sure, there have been abuses in the past -- and there will be abuses in the
future -- but perhaps you need to be reminded that we live in a world of spies,
thieves, cutthroats, skinheads, Mafiosi, and Scientologists, and often our
first and only line of defense is the FBI.  Sometimes they make mistakes, like
in the case of Steve Jackson Games, but do we help the situation by keeping the
FBI men ignorant?  I think ignorance is part of the reason why these abuses
sometimes occur.

In the particular case of censorship, this particular FBI man was interested in
how the net works and what relation it might have to criminal activity.  He
explicitly stated that the FBI was _not_ interested in censoring anyone's
speech.  When I told him about the recent discussion in alt.forgery, where
people were asking how we could discuss forgery without discussing the
committing of a crime, the reaction of the FBI man was "Well, you can write a
book about forgery."  My impression was that he was very sensitive to the
distinction between free speech and crime.  Just about the last words he said
to me were "It's a free country."

If the only thing keeping the net free of censorship is the ignorance of the
powers-that-be, that is not much of a defense at all.  In fact, I think it
would be naive to assume that potential censors aren't already aware of the
net's existence.  Uunet has revealed that they have sold compiled Usenet
traffic on tape to the FBI.  FCC men and telco security [people] are known to
read the Telecom Digest.  God only knows who reads RISKS.

Mark Thorson   (mmm@cup.portal.com)


Re: Risks of posting on the NET (Moonen, RISKS-11.72)

"Jim McLeod, ISSM, USAAPGSA" <jmcleod%apg-9@APG-9.APG.ARMY.MIL>
Tue, 28 May 91 9:21:08 EDT
If and when a relative or friend of someone who boasts of such antics is a
casuality of an airline bombing or other terrorist act, I hope he remember that
his actions probably helped the terrorists and that he might be an accessory to
murder of innocent men, women and children.  If what he said about NSA is true,
then did he ever stop to THINK that the time spent assessing phony "keywords"
can prevent the investigation of an actual terrorist plan to commit an
atrocity?  I am as concerned about privacy and computer risks as the next
person, but I get frosted reading such comments.  Such an individual would
probably be the first to blame "Big Brother" for not preventing the slaughter.
[Disclaimer implied]


Re: The RISKS of Posting to the Net (Moonen, RISKS-11.72)

<erspert@ATHENA.MIT.EDU>
Tue, 28 May 91 14:06:48 -0400
> Arghh. That's all we need now. Next thing, someone who says potentially
> dangerous words on the net, like say, ehh... blue box (Get that guys, BLUE
> BOX), or ehh... assassination of BUSH, will get a visit from our beloved Big
> Bro.

Actually, it has already happened.  A few years ago, I heard that
someone who had posted something to the net about assassinating the
president was visited by the FBI.

I suppose that this message will also end up in front of some government
worker.  :-(
                    Ellen Spertus


Re: The RISKS of Posting to the Net

Mike Olson <mao@postgres.Berkeley.EDU>
Tue, 28 May 91 11:29:08 PDT
In RISKS-11.72, Phil Agre <phila@cogs.sussex.ac.uk> takes mmm@cup.portal.com to
task for disclosing "dangerous" information about the Internet to an FBI agent.
While I sympathize with Agre's concerns, I think it's important to maintain
some perspective.  Censoring one another in our dealings with the government is
no different from having the government censor us directly.
                                                        Mike Olson, UC Berkeley


Replicated Errors

Robert McClenon <76476.337@compuserve.com>
27 May 91 22:53:40 EDT
I respectfully submit that Neil Rickert is completely and very seriously wrong
as to whether sendmail is primarily responsible for the replicated error
messages.  He would place the burden of preventing malicious software from
taking over networks primarily on network users and only secondarily on network
managers.  The moral burden of preventing malicious software of course falls
primarily on users and only secondarily on network managers.  Malicious acts
that take over networks by users are immoral.  The primary moral responsibility
for the Morris worm rests with Morris, and he was convicted of computer abuse.
But the primary pragmatic responsibility for preventing network abuse rests
with network managers, since there is always a risk of a malicious user.
Morris only exploited known vulnerabilities in sendmail and other software.

The message that Mark E. Davis disseminated was, because of the oddities of
sendmail, an accidental INTERNET WORM.  Anything that can be done accidentally
can be done INTENTIONALLY.  Network managers have a pragmatic and a moral
responsibility to prevent worms and other malicious programs and malicious
messages from taking over their networks.  To attempt to shift this
responsibility entirely to users overlooks the fact that not all users are
moral. Someone could have released a message with a malformed header on purpose
to flood the Internet with error messages.  The primary problem is not with the
user (Davis or anyone else).  It is with the software.  What Rickert says is a
robust programming practice is simply not that if networks have finite
capacity.  Fair is correct.  The error should be halted and sent back to the
user, or sent forward without comment, or corrected and sent forward.
Otherwise a WORM is possible.


Are fuzzy controls risky?

<HORN%HYDRA@sdi.polaroid.com>
Tue, 28 May 91 14:59 EST
Regarding the recently asked question about whether to trust a fuzzy logic
based controller for nuclear power plants.  I think the correct answer is that
they should be trusted as much (or as little) as you would trust any
controller.  There is no guarentee that a control system is stable or accurate.
This applies to both crisp and fuzzy controllers.  The mathematical theory for
robust control is better developed, but you must do the analysis before you
assume stability or accuracy.

Software engineers often make the mistake of skipping this step.  Many are
unaware that there is a large, well-developed branch of engineering called
system control theory.  Just because the control laws are precise does not mean
they will be accurate or stable.  Often the mathematics are inherently unstable
and no amount of precision in the software can overcome this.  The instability
is not a software problem (as assumed by many programmers); it is inherent in
the mathematics of the control laws.  These must be analyzed and shown to be
appropriate.

Fuzzy control stability is not as well developed a theory, but the mathematics
of fuzzy control are quite deterministic.  Stability and accuracy criteria may
be more work to evaluate.  The control problem is usually simplified
considerably for the fuzzy domain, so the resulting controller may be as easy
to analyze as the precise controller or perhaps even easier.

The nuclear power plant is an example of a situation where both may be
appropriate.  Consider the sub-problem of measuring the temperature
distribution within the core.  If you can assume thermal equilibrium, then a
scattering of temperature measurements, plus the knowledge of conductivity of
the components, plus the knowledge of the shape of the power distribution
allows you to compute the temperatures, energy flows, and power generation
throughout the entire core.  If you cannot assume thermal equilibrium (e.g.
startup and shutdown), you must also have the heat constants of the materials,
the prior temperature distribution history, and the power generation to compute
the temperature.  So without equilibrium, you need two previously unnecessary
inputs and one previously computed output must now be measured (somehow).
These added measurements may drive the control laws beyond the regime where
they can be analyzed, and they may also just be unavailable.  Power generation
distribution can be very tricky to measure directly.  It is possible that the
fuzzy solution is the alternative that remains mathematically tractable.

Then again, they may just be chasing the fad of the week.  I see a lot of
``fuzzy'' systems being invented for marketing and PR reasons.

Rob Horn       horn%hydra@polaroid.com


AT&T billing problem: "computer error"

Charles P Pfleeger <pfleeger@TIS.COM>
Tue, 28 May 91 15:58:11 -0400
I recently encountered an AT&T billing error on our phone bill, which the AT&T
office acknowledged as a computer error on their part.  There was a call from
DC to NY which showed no rate code (E for evening, D for day, N for
night/weekend), showed 15.5 in the "minutes" column (when all other entries in
that column were integers, and [here was the real clue] cost $16.00, when
similar (night rate) calls to NY were much less.  I have heard from one other
person who had similar problem.  AT&T cheerfully resolves these matters when
called, but I don't know if they are planning to adjust bills for people who
don't complain.


Caller ID in commercial applications

<web!roberson@igc.org>
Mon, 27 May 91 19:06:22 -0700
People have been warning that with automatic caller identification, all
sorts of strange and possibly undesirable cross-referencing will become
common. I have just run across my first reference to a mass-market
commercial program which will use caller identification in this manner.

The product is TeleCenter 2.0, developed by Northern Telecom. The
advicle [advertisement in the form of an article] in a recent MacWeek
[V5 N20, May 21/91, pg 27]  said, in part,

  "Features such as caller ID and TeleCenter's address book can be
  triggered from another application. For example, using caller ID, an
  order-entry application could automatically retrieve customers'
  addresses and buying preferences when they call to make an order."

Walter Roberson             roberson@web.apc.org

Please report problems with the web pages to the maintainer

Top