ARMS-D-Request@MIT-MC.ARPA (Moderator) (12/23/85)
Arms-Discussion Digest Sunday, December 22, 1985 11:47PM
Volume 5, Issue 68
Today's Topics:
Re: Phone Network Analogy
Re: Jastrow's claim
SDI Debate at Stanford
Re: Better Dead...
----------------------------------------------------------------------
Date: Sat, 21 Dec 85 19:25:05 est
From: Barry Shein <bzs%bostonu.csnet@CSNET-RELAY.ARPA>
Subject: Re: Phone Network Analogy
I find this analogy to the phone network hysterical, did anyone ever
actually interview the people who operated it? I had a friend who
worked on development at Bell and she would fill an evening with
humorous stories about its failures ("Whooops, there went New York
City!" was a typical punch-line.) This was NY/NJ in the mid 70's (is
this the same system the claim is being made about?) I remember she
mentioned that one reason the system seemed so reliable was that if
you got cut off at a pay-phone in Manhattan you had no way of knowing
that 25,000 other calls got dropped, it's just the kind of event
people don't put on the 6:00 news, you just curse and re-dial (by
which time they have switched in a backup system so things are almost
always okay real fast.)
So how can any of this apply to SDI? I surely hope the SDI supporters
have a few more examples besides this one, talk about the Emperor's
New Clothes! Then again, maybe it *is* analogous, if you get
vaporized you probably won't know that a few hundred thousand others
just got vaporized ("Whoops, there went New York City!")
-Barry Shein, Boston University
P.S. By "operated" I don't mean you dialed "0" to speak to her, I
mean she was a software engineer within Bell.
------------------------------
From: decwrl!decvax!utzoo!henry@ucbvax.berkeley.edu
Date: Sat, 21 Dec 85 22:31:50 est
Subject: Re: Jastrow's claim
Dave Schmitz writes:
> Is there any comparison between a relatively simple switching network
> and a battle management system? ...
Yeah, the switching system conceivably may be more complex than an SDI
battle-management system ever would be. There is nothing simple about
the worldwide telephone system, unless you consider only a tiny subset
of it. When you figure in the hardwired stuff as well as the software --
remember that the phone system functioned for many years before software
started to play any significant role in running it -- it is a computer
system (a highly specialized one) of complexity far exceeding anything else
ever built by man. Things like SAGE and the Internet are inconsequential
toys by comparison.
This is not to say that the experience necessarily carries over to SDI,
but let there be no illusions about relative complexity.
Henry Spencer @ U of Toronto Zoology
{allegra,ihnp4,linus,decvax}!utzoo!henry
------------------------------
Date: Sun, 22 Dec 85 12:38:47 pst
From: "Dave Caulkins; Computer Professionals for Social Responsibility; 415-322-3778" <cdp!caulkins@glacier>
Subject: SDI Debate at Stanford
Here's my account of what happened at the SDI debate held at Stanford on
19 Dec 85. These are my IMPRESSIONS of what occurred; the account is by
no means complete, exhaustive, or free from bias and/or error. I
believe that the event was video taped; if and when the tape becomes
available I recommend it.
The title of the event was "The Strategic Defense Initiative:
Can The Computing Requirements Be Met ?"
Panelists:
David Parnas, Professor of CS at U. of Victoria, resigned from
SDIO panel.
Richard Garwin, IBM Fellow, physicist, consultant to DoD
Major Simon Worden, Science Adviser to SDIO Director Lt. General
James Abrahamson
Richard Lipton, Professor of CS at Princeton, member of SDIO
panel (also ex-student of Parnas).
Moderator: Dr. Marvin Goldberger, President of Cal Tech
John McCarthy of Stanford made some opening remarks. He was
dubious that the evening's entertainment would have much
technical content.
Dr. Goldberger briefly outlined the background - the US has
about 11,000 strategic warheads, the Soviets have about 9,000.
He talked about Reagan's SDI speech, and quoted relevant
passages. He asked "Can it [SDI] work ?" He then said that
there are other questions which must be answered first:
"What is can ?", "What is it ?", and "What is work ?"
He hoped that the evening's discussion might shed some light.
He described another problem - the difficult transition from the present
offense-based strategic balance to a defense-based balance.
He outlined the rules for the debate - each of the four
panelists to have 20 minutes to state a position; then each of
the four to have 5 minutes for a rebuttal. This to be followed
by written questions submitted by the audience.
The first speaker was Maj. Worden. He made the follwoing
points:
In order for the SDIO research program to be judged a success,
and be considered for deployment, the system must meet a number of
strategic criteria:
1) It must be stable in a crisis
2) It must be 'arms control' stable - deployment of SDI must lead to
fewer nuclear weapons and less weapons-related tension.
It also must meet some technical requirements (first proposed by
Paul Nitze):
3) It must be cost-effective; costs-per-warhead-killed must be less
than the cost of any reasonable combination of offensive
countermeasures and/or building more offensive forces to saturate the
defense.
4) It must be robust, able to carry out its mission in the face
of component failures.
5) It must be survivable, able to carry out its mission after
direct attack on it by an opponent.
The present thinking of the SDI office is that battle management can be
done if each of the elements are independent.
At present offensive strategic warheads cost about $20 million to $30
million each, deployed [I presume this estimate allocates the entire
cost of an ICBM to the warheads]. They estimate that fast burn boosters
(complete the ICBM booster burn within the atmosphere to shorten the
burn time and allow atmospheric shielding) will raise the deployed cost
of warheads by a factor of 3, to $100 million.
The SDI Office now thinks that defense can have a cost advantage
over offense of about a factor of 10.
The present SDI program costs 1% of the total defense budget; surely it
is reasonable to spend this much on defense.
Richard Garwin spoke next.
He described the Star Wars system, illustrating its great complexity.
He went through many of the possible countermeasures, most of which have
been dealt with on ARMS-D. He described one that was new to me -
anti-simulation. The Soviets have lots of old boosters; site them on
the surface and fire them as decoys when the real ICBMs are launched.
Program the ICBMs to 'stagger' in flight, and otherwise display
signatures as similar to the decoys as possible. Put some real warheads
on some of the decoys, so the defense cannot ignore them even if it can
discriminate. Similarly during the deployment of warheads and
countermeasures from the MIRV bus - make the warheads simulate the
decoys and other countermeasures.
He said that SDI research costs more than all other government-funded
reserach.
Garwin proposed an alternate to Star Wars - negotiate with the Soviets
to cut back to 1,000 warheads on each side. This would require both
sides to abandon counterforce strategies. Spend about $1.5 billion/year
on ballistic missile defense research.
Lipton spoke next.
He gave several examples of scientific predictions of impossibility that
turned out to be badly mistaken.
He talked about the US banking system as an example of a system that has
never been tested as a system, is made up of a bunch of autonomous
parts, and works reasonably well.
Lipton said he didn't think the battle management system proposed by the
Fletcher Report could work, and that he agreed with many of Parnas's
criticisms of it. Lipton suggested that there be no Star Wars
management system. He proposed that the defense system consist of a
bunch of platforms, each with its own sensors, weapons, and computers.
Each platform would be autonomous, and each could have software written
by a different vendor (even one with software written by the KGB !).
There might be coordination between the platforms, but only on an
advisory basis. Each platform could be tested separately, and thus the
defense system would be testable. New platforms with new sensors,
weapons, and/or software could be added to the fleet at any time. A
good strategy might be to vary the number of platforms fully armed to
fire weapons as a function of the level of alert - only a few armed at
DEFCON 1, lots at DEFCON 3.
Parnas was next.
Lipton's autonomous platform scheme is no good. Even if you don't call
the meta-system of all the platforms a system, it still is one and has
to be dealt with. The defense problem is sufficiently difficult that
all possible information must be gathered and used; this will require
interaction between platforms.
Parnas went over the arguments he has made before at MIT and in the
papers he submitted as part of his resignation from the SDI Battle
Management Computing Panel.
Parnas pointed out that the argument for a 'layered' defense is only
valid if each layer is statistically independent of the other layers;
this assumption has not been proven and is probably false. Any decrease
in kill performance, especially in the first layers, can drastically
increase the number of penetrating warheads.
He summed it all up in three assertions:
We will never be able to trust the SDI software; if we cannot trust it
we will act as if it isn't there and continue to rely on offensive
weapons and MAD.
Our opponents will have to make decisions on the assumption that the SDI
system will work; the results of these decisions will be bad for us.
Human beings cannot build software that is right before it's tested.
Rebuttals (each panelist got 2 minutes)
[my notes and recollections are particularly sketchy here]
Lipton:
Nothing is perfect.
Can 'a thing' that is reliable be built ?
The defensive system needs a lot of independent, imperfect pieces.
You may be able to give up multiple 'views' of a sensor-detected object
if teraflop computing power is available on each platform.
Parnas:
As an illustration of system-foiling by clever people, he told an
anecdote about a banking fraud in which a $38,000 check was generated
with the magnetic code for one bank but the name and address of another.
The check ping-ponged between the two banks for a long time before
humans discovered what was happening; by then the $38,0000 was long
gone.
Lipton said yes, but the whole banking system didn't collapse.
Question Period
Worden:
We welcome a Soviet SDI if done in coordination with the U.S.
SDI sensors can be used to locate cruise missiles; this is the hard
part and once done other non-SDI means could be used to shoot them
down.
Parnas:
There is probably enough time for human intervention in the overall
decision to fire SDI weapons or not.
There was a good deal of discussion of pop-up weapons; Garwin took the
position that fast-burn boosters made any pop-up scheme impractical.
------------------------------
Subject: Re: Better Dead...
Date: 22 Dec 85 14:36:54 PST (Sun)
From: Tim Shimeall <tim@ICSD.UCI.EDU>
I had indeed wrongly attributed that quote to you [LIN]. I apologise.
Tim
------------------------------
End of Arms-Discussion Digest
*****************************