Lewiston Tribune/Friday, October 11, 1991, page 6C Associated Press FEDERAL WAY, Wash. -- King County Police confounded by a typographical error mistakenly descended on the home of Terry and Dean Krussel this week -- for the second time this year. At least this time they didn't break the door down. When the officers from the narcotics unit raided the Krussel home in May, they kicked in the door, ordered Terry Krussel, 57, to get down on the floor and held her at gunpoint while they searched the house. County officials replaced the door at a cost of $2000 and apologized profusely. When the Krussels got a letter from the county prosecutor's office on Sept. 11, addressed to the person officers had sought in the May raid, they worried that their address was still on file as a den of iniquity and dangerous drugs. King County police scrambled to delete their address from the department's computer files, and deputy prosecutor Judith Callahan assured the Krussels in a Sept. 17 letter of the county's good intentions. "Our office is truely concerned that Mr. and Mrs. Krussel not feel that they are victims of county bureaucracy," she wrote. Unfortunately, the Krussels' address remained in the drug dealer's file -- and that's what the officers pursuing the dealer Tuesday night were working from. The officers didn't leave until Dean Krussel showed them Callahan's letter. "This thing just won't go away," he said after the couple's latest run-in with King County's finest.
>The debate is international in applicability. However, U.S. policy on encryption appears most severe, so I urge a U.S. National debate to begin the dialog, and start with some questions. I agree with Clark that debate is indicated. There is no proper forum for this debate. The present policy has ancient origins. They are older than the Cold War, though the Cold War has been used to justify them since the National Security Act of 1947. The current policy dates from the Great War and was placed in law without public debate in 1943. That law, passed in war time, has been used since to suppress any further debate. >Do we gain more by strengthening our commercial competitiveness and products, upon which the military is increasingly dependent, than we lose by permitting international commonality in cryptographic services, which may weaken military capabilities? While it is difficult to state the issue, proper debate requires that it be stated clearly. I do not think that Clark's question properly frames it. I think that the issue is more one of the trust and confidence required for commerce than it is one of "competitiveness." This country needs trade. The most efficient way to mediate trade in the modern world is electronically. Trust and confidence in electronically mediated trade requires secret codes which both parties can trust. That is one interest. I submit that it is far more compelling than mere "competitiveness." I also understand the contending issue differently. Rather than relative "military capability," the issue is one of the cost of intelligence gathering. Even in a peaceful world, security requires that we gather intelligence. Prudence suggests that we gather it about everyone, not simply "adversaries," but everyone. History screams that any political instability causes people to choose sides. Therefore, it behooves us to know as much as we can about what is going on in the world. If the ether begins to fill with "random appearing" data, the cost of intelligence gathering will rise as a geometric function of the quantity of that data. Therefore, the second interest is to discourage that data to the extent that we can. It is not simply one of effectiveness; we cannot hope to discourage all use of secret codes. Rather it is one of efficiency; how much can we discourage and at what price. Neither of these interests is trivial. Each is worth defending. They do conflict. To date they have been debated only in secret proceedings. I am concerned that in those debates, the latter interest has prevailed and that the former may not have been properly appreciated. I do not believe that either interest will be seriously compromised by a more public debate. William Hugh Murray, Executive Consultant, Information System Security 21 Locust Avenue, Suite 2D, New Canaan, Connecticut 06840 203 966 4769
One note of caution here... > Criteria for secure time-sharing systems will not "make it" in the nineties, but it is not clear that we know enough to write evaluation criteria for networks, data bases or applications... I think I see the Wheel Of Reincarnation operating here in several ways. Time-sharing systems are passe', but everyone is busy rediscovering the same old issues in the context of networks, databases, etc. What, exactly, is the fundamental difference between a time-sharing system and (say) a heterogeneous network? Answer: there isn't one, unless you insist on thinking of time-sharing systems in terms of a narrow stereotype that has never described all time-sharing systems. (As a case in point, note that the Plan Nine experimental operating system at Bell Labs is aimed specifically at making a heterogeneous network look pretty much like a time-sharing system. They're succeeding fairly well.) What, exactly, is the difference between the access controls enforced by a shared database and the ones enforced by a time-sharing kernel? Answer: while there is a different flavor to some of it, the problems and solutions are often very similar. And so on. "Those who do not remember history are condemned to repeat it." If we continue to discard past experience with multi-user systems as obsolete, we will continue to rediscover issues and make the same old mistakes when building new multi-user systems. Criteria for secure time-sharing systems deserve very careful examination, as much of that experience should be applicable to networks, databases, applications, etc., given some caution in the presence of shifts in the underlying concepts. Henry Spencer at U of Toronto Zoology
Ah, but the people who wrote the Orange Book (TCSEC) years ago were thinking not in terms of generic functionality for trusted distributed systems, but primarily in terms of isolated-system security kernels. They wrote the criteria in an overly-specific manner that makes the applicability to networks and distributed systems very difficult/uncharted/unclear/... Nevertheless, the Red Book tries... See also the European ITSEC. But in principle any sensible operating system concept could be distributed in a nice clean invisible way; in practice there are LOTS OF PROBLEMS, some of which are indeed different from the old ones (such as distributed authentication).
>The USAF/NASA Advanced Fighter Technology Integration test aircraft is doing flight evaluations of a system to help pilots cope with disorientation: push a button on the stick and the computer automatically brings the aircraft back to level flight. As an active aerobatic pilot, I've had the experience several times of complete disorientation, the horizon cannot be interpreted or seen and the G-forces acting on the body lead to incorrect conclusions regarding the attitude of the aircraft. Although a mechanical device to recover from an unnatural flight situation would be of immense benefit, the process of achieving level-flight from a given spatial orientation can be quite complex, involving judgements regarding G-Forces, rudder & aileron coordination (or dis-ordination in some cases), airspeed (both indicated & true), aircraft red-line and stall characteristics, etc. These factors can for the most part be vectorized into a given computer system/program assuming that *ALL* of the sensors are functioning correctly; the consequences of going over red-line and getting flutter due to a partially blocked Pitot-tube or going into an unrecoverable stall because the aircraft isn't balanced correctly on this flight (perhaps the cargo shifted) and the recovery-software wasn't informed or stalling because there is icing and the stall-warning is out of function are devastating. The risks inherent in such a system would be outweighed by the benefits when an emergency situation occurs assuming the pilot has no recourse; but the knowledge that the aircraft is equipped with "a device which will get me out of any situation" might make a pilot take more risks and thus induce exactly that situation where the system must be used; somewhat akin to the RISKS article about the warning signs for the Virginia drawbridge.
The AFTI/F-16 is a completely instrumented airplane and has several accelerometer packages. Level flight just turns into setting a_x and a_y to zero, with a_n = -a_z = 1.0 g. You can couple in the rate gyros and set p, q, and r to zero too. This is a pretty simple little feedback system. >I can't see how this device is better than your basically-trained >IFR pilot, and it may be worse (mortal failures under strange >instrument failure modes). Quinine, in the form of tonic water, doesn't give the accelerometer package vertigo like it does the pilot's vestibular system (being discussed in rec.aviation right now). Accelerometers don't get the leans, either. Actually all F-16s have similar accelerometer and rate gyro packages, the AFTI/F-16's are just tied to the instrumentation package as well. Modern fighters are somewhat more heavily instrumented than are general aviation aircraft (which, by the signature, is what the poster, a private pilot, is familiar with). The system was first proposed to deal with GLOC (g-induced loss of consciousness). The F-16 is notorious for having such a high instantaneous rate of g onset that pilots in combat are at risk of GLOC. The question is really whether this system is better than an unconscious pilot. (To further tie this to a thread in sci.military, the F-20 had the same high g onset rate and many people believe that GLOC led to at least one of the prototype crashes.) Mary Shafer DoD #0362 NASA Ames Dryden Flight Research Facility, Edwards, CA email@example.com firstname.lastname@example.org
All technical issues aside, the first (obvious) question that comes to mind is: Who gets to design, build, program, install, verify and maintain this system? If all countries got along well enough to settle that question, the whole issue would become moot (perhaps that's the real idea). But another very real question is: Would it really be an improvement over the existing situation? Maybe I'm missing something, but wouldn't it just make Mutually Assured Destruction even more "mutually assured". Unless, of course, the idea is to give the victim enough time to completely destroy the attacker at the outset. In that case, it would be as "realistic" (and a lot cheaper) to get countries to agree to just blow themselves up if they ever get angry. What I really don't understand is, if (a huge "if") it WERE possible to establish central, tamper-proof control over ALL countries' abilities to launch ALL nuclear weapons (as the article suggests), why not go the one (IMHO small) step further and make it impossible to launch them at all? ...maybe through the use of a "beneficial virus" (just kidding). Personally, I think it would be very sad if the world could achieve the sort of trust and cooperation necessary to implement this system, and not not manage to do away with the things entirely. Eric Prebys, CERN, Geneva, Switzerland
I can't vouch for the details of the arguments, but this is a good example of trying to decide a scope of solution. Is it more important to maintain a given system in its own cocoon or take the risk of change in order to get the benefits of what, over a decade, has emerged as a standard. We can argue the technical benefits (and I would think that 10 years of change has produced some improvements, though nowhere near as much as it might have) but there are larger issues such as switching into a more cost effective/price competitive market. There is also the benefit of standardization in terms of being able to take advantage of common knowledge and tools. Risks are necessary part of evolution. It is important to be aware when one is taking a risk and the consequences and not be naive. But not taking a risk can be a bigger risk. Again, I claim no knowledge or insight about this particular instance and I'll admit to a bias in favor of rampaging PC's.
I used to use UNIVAC 1100 series computers too. However, I would like to cast some doubt on one of the statements, and refute another made under this topic in RISKS-12.47. First of all, there were other companies who had well developed realtime software processing more than 12 years ago, although perhaps not on a processor the size of a, say, UNIVAC 1108. For example, one machine and operating system that comes to mind is the PDP-11 running RT. Then there is the VAX... And speaking of the VAX, it is a system sold by another vendor (Digital Equipment Corporation) which has as large a range of compatible processing power as the UNISYS 1100 series, if not more. Since I am a software specialist, I'll spare the sales pitch... Doug Burke, Senior Software Specialist, Digital Equipment (Malaysia),
>Unisys 1100-series equipment, from the smallest (2200/100, desk sized small business system) to the largest (2200/600, big mainframe), runs the same software across the entire line with NO modifications required. Such a large range of compatible processing power is unavailable from any other vendor (the Unisys A-series has a somewhat wider range). I must take exception with this. The VAX family processor will faithfully execute programming which makes no installation dependant call provided that the VMS linker was used to link it, and that the VMS executive is at the same release point or is a later release. I believe DEC is not owned by Unisys. Also, will the Unisys equipment take the executing image of the code, or must the source be recompiled? The VAX family processor, for example, executes the exact same executive no matter what model it runs on. Actually, I do recall one release of VMS where that was not the case, but then DEC fixed it anyway. Guy Sherr, Lab Configuration Mgr, MCI Reston, VA email@example.com
Perhaps our moderator, Peter Neumann, should just keep a copy of this on hand and reissue it as needed. This is long, but needed periodically. (Maybe Peter should add a brief version to the masthead of RISKS!) The real RISK in computer system design is NOT human error. It is designers who are content to blame human error and thereby wash their hands of responsibility. in RISKS The ATT failure The truck driver and bridge In Aeronautics Digest 3.22 (Oct. 10, 1991) Traffic collision avoidance system failures: the Federal Aviation Administration (FAA) ordered a shutdown of 200 of the 700 units that had been installed. The 200 systems were seeing phantom aircraft and instructing pilots to evade planes that simply were not there. "We had a simple human error where an engineer misclassified the changes in the software" Human error is almost always a result of system and design error. It has to be taken account of in the design and in the work procedures. Lots of people in Risks have proposed design procedures that will help. Even the manufacturer of the TCAS system (in the last incident above) said: . To prevent similar omissions, Collins now requires that a committee of . software engineers review changes before a program is released. "More than . one pair of eyes must review these things and make a decision" That will not guarantee correctness (if, for example, the specifications are incomplete or inappropriate -- as they almost always are -- the committee will simply verify that the program meets the wrong specifications) but it will help. Committees are also subject to various kinds of group decision processes that sometimes propogate errors. It is a first step, but it still does not indicate that the designers are sensitive to the nature of error and will take design pains to avoid it. Example: if only the truck driver had been attentive, the accident would not have happened. True. But also if only the signs had been working, or if the procedures required traffic to stop elsewhere, or if only the drawbridge hadn't been raised. In any accident, there are always dozens of "if onlys". NO HUMAN IS 100% ATTENTIVE. Designers assume perfect human attention, which is fallacious. (My restatement is that humans are excellent at switching attention among competing demands. Alas, the demands of modern technology are not always compatible with the evolutionary structure of the human.) The design error is assuming inappropriate properties to humans and assuming they can perform in ways that are foreign and unnatural -- truly, biologically determined, "hard-wired," unnatural. We design to allow equipment to work in the face of noise and even component failure, certainly in the face of out-of-tolerance components. We should do the same for people. It is no excuse to blame training, attention, attitude, or "human nature." These things happen so much that they have to be designed for. And we even know how to do so. The real problem is the attitude of the design community, even among those who read RISKS. The other problem is the training of the design community: engineering and computer science departments train technology, program verification, and the like. No expertise in human and social issues. Computer scientists cannot turn overnight into social scientists, nor should they. The design of systems for people requires design teams consisting of computer scientists, cognitive and social scientists, (and representatives from the user community). Technology alone cannot provide the answers when we deal with human activities. "What has this to do with computer science? Nothing, directly, but indirectly it means a lot. The same computer that makes so much possible, also sets up the conditions for human error. And if this is not understood, the systems will fail, and the failure will be blamed on "the computer" or even on "those computer programmers and scientists." (rephrased from Norman, in press) SEE: Perrow, C. (1984). Normal accidents. New York: Basic Books. Norman, D. A. (1990). Commentary: Human error and the design of computer systems. Communications of the ACM, 33, 4-7. Norman, D. A. (in press, 1991). Collaborative computing: Collaboration first, computing second. Communications of the ACM, 34 Donald A. Norman, Department of Cognitive Science, University of California, San Diego La Jolla, CA 92093-0515 firstname.lastname@example.org
Some people seem to want to blame human weaknesses for the AT&T failure, other people seem to want to blame the technology. What I havan't seen anyone point out is that, every time AT&T (or most other people) does something to "improve" their system, they end up more and more centralized. Actually, that's partly what I was trying to point out in my earlier post. When you install the same software (do they still call them "generics" inside Bell?) everywhere, you have an implicit single-point-of-failure across the whole network. Yes, when they route too much through a physical single point of failure, that's bad, and they know it (or should). But it appears to me that - the historical system availability target of 2 hrs outage in 40 years is no longer being met, even though it once was with much lower tech hardware - the reason may be related to this implicit single-point-of-failure not being made explicit in the way the code is written, or the development project is run. Perhaps the attitude they took with the Space Shuttle computers needs to be transferred to the phone company. Yes, do a superlative job in programming the four on-board computers, write it to the most exacting specifications, then test the heck out of the code. But oh-by-the-way, here's a fifth computer with completely alien hardware AND SOFTWARE in order to obviate any implicit, unanticipated, yet catastrophic single-point-of-failure modes. I don't know that this solution can or should be adopted wholesale; it's the frame-of-mind that the shuttle designers had that there is this class of problems that appears to be lacking in the current design of the phone system. Bob Colwell, Intel Corp. JF1-19, 5200 NE Elam Young Parkway, Hillsboro, Oregon 97124 email@example.com 503-696-4550
I think we're in danger of missing a key element in the AT&T outage. Yes, the technicians were lax (if understandably so); yes, AT&T had routed too much stuff through the one switch w/o any backup path (which I think was the chief screwup); yes, the alarm system was inadequate (which AT&T has promised to address). But the real problem is that the power system and its alarm system were designed under the assumption (now vitiated) that technicians would be there to supervise it. Recall that AT&T says the rectifier failure was discovered only when a technician happened upon an alarm registering at a location (away from the power equipment) which was not ordinarily manned. That location would have been manned before AT&T riffed many of its technicians. The alarm system was not adequate to alert the present human supervisory regime-- perhaps the old technicians should have been kept on the job until AFTER the promised new alarm system was installed? Beefing up drills as suggested by some is an inadequate response to the design-constraint/ reality gap evident in the description of the AT&T setup. As hard as it is to get the humans to meet the needs of the system, or the system to meet the needs of the humans, if we don't try to match them at the interface as best we can, failure is certain. Building a machine which needs a supervisor, then firing that supervisor and expecting all to be well is foolish. Mark Seecof <firstname.lastname@example.org>
>This seems equivalent to the question of how much override a pilot of a fly-by-computer airplane should be able to exert; when the flight computer refuses to pull too many G's because the wings may overstress, but the pilot knows he'll hit a mountain otherwise, it's a bit clearer who should outrank whom. Perhaps in that specific case, but in the general case it's not that clear. I haven't studied the statistics (if anyone even has any along these lines), but what if the data show that more people die because the crews override when they shouldn't than because they can't/don't override when they should. We have already had a couple of Airbus losses in which a suspected cause is the crew inappropriately overriding the flight computer and riding the aircraft into the ground (e.g. Toulouse airshow). Have we lost any because the crew failed to override? Have we lost any other air transport types because of inability to override? Speaking as a pilot myself, emotionally, I always want to have total authority of the craft, but if statistically I am more likely to live longer by not having (or at least not exercising) that authority, my preference is not completely obvious. Perhaps the "automatic | manual" override switches need to have big legends above those descriptions, stating "PROBABLY | USUALLY SURVIVE | PERISH " Bob Niland, 3404 East Harmony Road, Ft Collins CO 80525-9599 Internet: rjn@FC.HP.COM UUCP: [hplabs|hpfcse]!hpfcrjn!rjn
We could make the humans the prime operators, and use the computers as a back-up. This preserves the motivation - noone wants to be caught making mistakes - and gives many of the desired benefits. Of course, we still cannot predict the reliability of the overall system, but that's another problem :-(
Although 00-55 is an interim standard, it seems that there is real progress towards its development and eventual adoption. About a year ago, we produced a VDM specification of the safety requirements for an ammunition control system (ACS) which is used by the Directorate of the Proof and Experimental Establishment of the MOD for managing the ammunition holdings of some ranges. I understand that the appropriate MOD authority intends to issue our specifi- cation as a part of the Operational Requirements draft for the next generation of the system. I believe that the intention is to provide an improved statement of the safety requirements during the tendering process. Although, this is a long way from full application of 00-55/56, it is certainly an encouraging and a very welcome step in that direction. We have a technical report for anyone who is interested. Victoria Stavridou PS. If you want to followup this topic, please email me direct because our news server is down at the moment.
The latest Hammacher Schlemmer catalog advertises a "Voice-Changing Telephone", that uses digital signal processing technology to realistically alter the sound of the user's voice, even changing male speech to female, child to adult and vice-versa, to completely disguise identities and discourage unwanted calls. Perfect for people living alone or children at home by themselves . . . Yes, and perfect also for abusive or threatening telephone calls, imposters' scams, and sexual harassment. Even if used in the way that the advertisement suggests, some peculiar scenarios emerge. E.g., Deep Male Voice: Mommy and Daddy aren't home right now.
Please report problems with the web pages to the maintainer