On October 29 I was ensnared by a design flaw in the security system of the parking garage of the San Francisco airport. The same system is undoubtedly used at other airports. I had returned from a trip that morning and paid $88 for 8 days. I returned the same evening to fetch my wife, who returned from a trip. On presenting my ticket, the attendant said the computer said I owed $99, rather than the $1 I was expecting to pay. Guards appeared and instructed us to go to the garage office to discuss the matter with garage officials. After some discussion the garage official allowed us to leave, paying $1. Here's what happened. The garage security system is set up to prevent a customer from obtaining a second entry ticket on his return and thereby underpaying. When I entered the garage 8 days ago, a video camera recorded my license number and associated it the parking ticket dispensed by the machine; the resulting license-ticket entry record was placed in a database later that night. When I checked out, an exit record of my license-ticket pair was made, and scheduled to cancel the entry record during the late night computer run. However, when I attempted to check out the second time, I had not been there long enough for a new license-ticket entry record to be placed in the database; accordingly the standard database check found the still unpurged record of my first entry and computed that I owed for 9 days. The garage official apologized for the inconvenience and said the system is needed to prevent fraud and occasionally someone stumbles into a false alarm. There is no interest in changing the system or posting notices warning customers to keep their receipts if they return to the garage a second time in one day.
From an article in MODERN RAILWAYS, September 1987, by Roger Ford, on the opening of the Docklands Light Railway (DLR) in London. Submitted to RISKS (and slightly edited) by Mark Brader: Ironically, the 'problems' the daily press reported at the Royal Opening were caused by the automatic system working properly. The royal train (number E2R — a nice touch that) was timetabled to leave Island Gardens station at 15:30. As the royal party was early, the control room dispatched the train manually rather than keep Her Majesty waiting for five minutes. This meant overriding the computer, which was operating in regulated mode. Unfortunately, computers are less sensitive to royal protocol, and when E2R arrived at Mud Chute station [!] five minutes early, it was given a "dwell time" of several minutes to bring it in line with the timetable. If you happen to be on board a stationary train with your Sovereign, two minutes is a very long time, so the train captain reverted to manual ... One of the golden rules for bodyguards is that you leave a vehicle or building first. As the train rolled to a halt in Poplar station, a security man used the emergency exit so he could get out first. Not only did this stop the train by interrupting the door-safety interlock circuit, it stopped it short of the docking beacon. Unless the train receives a message from the beacon it does not know it is in a station and the doors won't open. I was standing next to the manager of the doorgear suppliers at the time and can vouch for the fact that time also stretches agonisingly when there is the possibility that Her Majesty is being isolated from her loyal subjects in Docklands by your product! In the light of this, I wonder whether anyone has thought to give the police and counter-terrorist organisations a briefing on the features of automatic railway operation.
From: Rob van Hoboken +31 15 78-3813 RCOPROB at HDETUD1 I found the following gem in issue 10 (July 1987) of MVS update, a publication of Xephon aimed at the MVS systems programmer. > Dynamically making programs APF authorized > > The following procedure should work under any release of MVS. > > The standard way of making progrmas APF authorised is to link them into an > authorised library specified in the SYS1.PARMLIB member IEAAPFnn with the > link-edit attibute of AC=1. Changes to such a library specification will > require a re-IPL. > If authorised programs are to be executed under TSO, they must be put into > a table in the Link Pack Area: TSO commands go into table IKJEFTE2 and called > programs go into IKJEFTE8. Any changes to these tables require a re-IPL > with the CLP option to make them effective. > It is, therefore, convenient to be able to turn on and off the APF > authorisation dynamically for any program that does not have the AC=1 > attribute, in any library upon request. > Just be aware that this can be a security exposure. > The solution: > The method is to have a user SVC that sets or removes the APF bit in the > control block JSCB. This SVC is probably well known but it is, together > with the following two macros, a pre-requisite to many homegrown functions > that help to make life easier. > The following SVC can be enhanced by adding installation dependent security > checks: > * authorisation svc 235 type 4 > * r0 = 1 turns on auth > * r0 ne 1 turns off auth > * > code follows Personally I would rather omit the name of the author to protect the guilty (but most of all his company), but since there was a copyright statement on the publication: (C) Nils Plum, Systems Programmer (Denmark). In effect this persons describes a hole in his installation, and proudly tells us that many of his tools depend on it. Probably (I've seen it in several installations) the "installation dependent security checks" were removed at some time to make "all those goodies available to us users". The worst part is that a lot of software companies ship out programs that actually need these kind of trapdoors to function at all. I have written to some of these companies with a description of the problem, proof of its existence and a work around. I got laughed at, made rediculous and told to not to spread the word. One of these people (a real big company!!!) even told me: (direct quote!) "it must be safe, even the CIA uses it" Can anyone help me to a userid + phone number on either Mr. Plum's installation or the CIA? Rob van Hoboken, Delft University of Technology, Computing Center
The new release (4.1) to the public domain spreadsheet "sc" has a noncompatible change that, along with an overlapping windows Sun environment and an ambiguous (but fairly standard) naming convention, conspired to destroy a database. I maintain a database of members of a Spring Water Cooperative at my workplace in North San Jose. Each member contributes $3 a month to enjoy unlimited use of a bottled water cooler. The record of accounts is kept in a database maintained by sc. The entry for a member for any particular month (say, [K1]) is the minimum of the total_amount_contributed [A7] minus the sum of payments so far [@sum(d7:J7)] and the current ammount due [K0]. As a sc.3.1 expression, [@min(A7-@sum(D7:J7),K0)]. This morning our /usr/local manager installed sc.4.1 naming it /usr/local/bin/sc (simply replacing its predecessor). Soon after, a member gave me his dues for the month. I opened sc in a window that had its right half hidden by an overlapping window, yet I had access to the total_amount_contributed column. Thus I did not notice that the right half of the spreadsheet was blank! I updated his account and saved the database. What I didn't realize is that sc.4.1 didn't recognize this usage of @min(), as this had changed, and when it saved the database, it simply and quietly ignored (read: deleted) all of the entries which it didn't "understand". Thank Zippy for disk-to-paper dumps! -fen P.S. I am sorry to see this useful function disappear. The README states that the "range" function should be used in its stead, but I havn't yet figured out how. Old and new expressions for table entries appear below: old: @min(A7-@sum(D7:J7),K0) new: A7-@sum(D7:J7)<K0?A7-@sum(D7:J7):K0 Fen Labalme, Megatest Corp, VLSI Systems Division, 880 Fox Lane, San Jose, CA 95131 (408) 437-9700 x3382 "megatest!fen"@riacs.ARPA UUCP: ucbvax!sun!megatest!fen
if anyone would like detailed knowledge of the mu-2 accident rate, i can look up some analyses i have. please send me mail. roughly, there have been a number of unexplained `uncontrolled descents into terrain' with the mu-2. of the order of a dozen, mostly with experienced pilots, although not with a lot of mu-2 experience. some others that have been explained concern the proper operation of the autopilot. autopilots with altitude hold can enter an unstable feedback loop. e.g. the plane is trimmed to hold altitude, and deviates, say down. the pilot pulls up, whereupon the autopilot senses control pressure and commands down to counteract the pull-up. the pilot pulls harder, exacerbating the problem. the mu-2 is a very clean aircraft and can exceed `never-exceed' speed very fast in a dive from cruise speed. faster than about 115 per cent of this speed, and the aircraft starts to break. speculation is that the pilots don't figure out the problem before they reach these critical speeds. other speculation is that there is a failure mode of runaway nose-down trim caused by the autopilot. even other speculation is that control of the aircraft is coming up against human-factors issues that were (and still are) poorly understood. it's clear that many of the accidents are pilot-induced, as with many very-high-performance planes. peter ladkin, email@example.com [Late word from Nancy Leveson suggests that the equipment in question is analog rather than digital, but that is still quite computer related... PGN]
In RISKS 5:51, Joe Morris commented on conflicting alarms in a 727 accident. My aero engineering days are a few years back, but I seem to recall that a stall warning caused by high angle of attack (which translates to a function of [low] indicated airspeed and dynamic load, the product of weight and G-loading, assumed to be close to 1 for an airliner in cruise condition) and overspeed warning caused by Mach number limitations (it is the Mach buffet that is the problem at that point, not the dynamic pressure) occur together at that point in the flight regime so aptly named the "coffin corner". That point, unfortunately, is also the point of maximum specific range, is it not? (Which is why all airlines always try to fly as close to it as possible.) Several points seem worth noting: (1) Airline flight crews know about the coffin corner, they fly close to it all the time, do they not? The presence of the two alarms should not be considered as contradictory; they are both correct and indicate an unambiguous situation for which recovery procedures are (or should be) well known. (2) As to whether the two-alarm condition is confusing, the presence of two alarms that must be interpreted by the human operators (as opposed to a single, "coffin corner" alarm) is a function of the use of analog-mechanical alarm systems. The stall warning system operates (I presume) off of angle of attack; the overspeed warning off of Mach number (pitot differential and temperature). Neither system is connected to the other, the design would be cumbersome, expensive, and risk-inducing. Interpretation of the two-alarm condition is properly (for the analog-mechanical case) left to the human being. The introduction of digital, autopilots offers a chance to improve the situation somewhat. Instead of merely duplicating the set of alarms provided by the older technology devices, good digital system design would add the logic to check for both conditions and then generate a new, unambiguous, "coffin corner" alarm. What is cumbersome and risky for a mechanical system is much easier and hence perhaps appropriate for digital technology. The use of digital computers to detect, interpret, and indicate conditions caused by the interaction of multiple factors is a chance to reduce risk.
Matt Jaffe's comments are well taken, but I would like to add a few closing notes: o His comments about "coffin corner" are correct. From what I've read (no, I'm not a heavy-iron pilot) the best efficiency can generally be found where the Mach buffet meets the stall warning. One mark of a good pilot is the ability to find the best compromise between performance and safety margin. o The 727 crash I referred to, however, involved a *false* high-speed warning. The cockpit instruments indicated an airspeed of 420 kt at 24,800 feet msl with a rate of climb of 6,500 feet per minute. (not 30,000 feet...my error) This is far above the performance possible for the aircraft. The readings were consistent with the pitot heads becoming blocked as the aircraft climbed through 16,000 feet. o Finally, the integration of various sources of data to produce situation- specific alarms is not only a good idea, but is being done in various implementations already. (I assume that someone is working on a "coffin corner" warning; I'm not in a position to routinely see such stuff.) the problem is that regardless of the way in which data is processed, the GIGO principle still applies, and the sensors are of necessity still mechanical analog devices. If the primary data source is lying, the user of the data may detect that conflicting information is being received, but it's not always possible to determine which of several sources has failed. The Air Florida crash in Washington is another example of exactly this kind of situation. Oh yes...from time to time PGN asks for specific citations of incidents. The accident I was citing was Northwest Airlines, Boeing 727-251, N274US, near Teiells, New York, 1 December 1974.
New Method to Protect Privacy of Computerized Data is Patented, By STACY V. JONES, c. 1987 N.Y. Times News Service, 31 Oct 1987. WASHINGTON - A professor of computer science at a Virginia college has invented a new method of protecting the privacy of computerized data. He was granted a patent this week for a cryptographic system that he describes as much less time consuming than present methods. Professor Hito Asai of Christopher Newport College in Newport News received patent 4,703,503 for what he describes as simple mathematic computation, technically known as Vector Boolean algebra. The computer data is enciphered with a long numerical key. According to Asai, an eavesdropper who attacks the enciphered text would face a time-consuming process. He plans to offer licenses to computer and communications manufacturers. [It may take even less time for the cracker to break, if the long key is stored or transmitted in the clear... Simple numerical algorithms may also be subject to inversion. But this is certainly worth investigating. PGN]
I used to work on an index arbitrage trading desk and I downloaded and executed the baskets of stock orders that Mr. Nelson mentions in the Wall St. Article of Oct. 13, cited a few issues back. He relates a horror story of someone pushing a button over and over after not getting any response and buying $100M instead of $25M worth of stock — I heard this almost two years ago. Supposedly, the culprit was his printer being offline, and not printing the orders as they were sent down to the exchange. Currently, software distributed to brokerage houses by the NYSE has safeguards against this, such as ignoring input until transmission of a set of orders is complete. (Not everyone uses this software, though.) We've also seen share volumes of unheard of proportions. One statistic on TV was that there were more shares traded last week than in all of 1965. One explanation for this is the size of the orders allowed on the exchange's computer systems. Until last December, an order through DOT (the system which handles "market" orders) was limited to 2000 shares. Since then, that limit has been raised to 30,000. With the ability to transmit 100 orders per minute, one PC AT has the capability to buy or sell up to 3 million shares per minute. And there are many systems of this type on the street. (Of course, that does not mean that these machines are continuously in use or always dealing in volumes of 30,000 shs/order.) Are these trades being done completely without human intervention? The answer is no. The program trading most often referred to is index arbitrage. Arbitrage takes advantage off price discrepancies of the same good in different markets, and involves buying the underpriced one and selling the overpriced one. The most common index arbitrage is done on the S&P 500 and its corresponding futures. The futures market at the Chicago Mercantile Exchange still uses open outcry (a la Trading Places) to do business. This means that, although the computer screen may tell you that now is the time to buy stocks and sell futures, you'd better make sure your guy in Chicago did his job and got you a good price on the futures before you push your button and buy $x million worth of stock. There's substantial monetary risk involved, and most (if not all) traders would not surrender that decision to a computer. The portfolio insurance mentioned a few issues ago deals only in futures, which again means trading via open outcry, and not with computers. Computers are used in this strategy for modeling purposes, to tell the portfolio manager how many futures he should sell to hedge his market exposure. The only programs I've heard of that may possibly do transactions strictly by computer are AI programs that try to anticipate very short-term trends in the market and buy or sell just before the market moves in that direction. I have not seen these, and don't know how well they work, how widely they are used, or how "automatic" they are. As far as I can see, it is just the *ability* to move large amounts of money in and out of the market very quickly that has changed things. Computers are still not decision-makers, although they do provide real- time data and much quicker reaction time to that data. But alot can happen in the five minutes it would take to buy or sell all 500 stocks in the S&P 500. It is the trader who makes the judgement whether the transaction could be profitable or not. It should be noted that the DOT system was not allowed to be used for index arbitrage programs for the rest of the week after the big drop last Monday (10/19), yet the daily volume was still much higher than had ever been seen before. In addition, the President of the Chicago Mercantile Exchange said that program trading accounted for only about 10% of the volume on the NYSE on 10/19. (New York Times, 10/28, p.D11) I would take this to mean that the programs were not the driving force in the market moves, and that it was real selling. It is possible to do these transactions without computers, but harder to make them profitable because of the extra time involved when human agents are used, and therefore much less likely that they occurred when using the computers was not allowed. The time windows for the existence of profitable market spreads are often as short as 30 seconds. One argument defending program trading's market impact on stock prices is that it reflects real selling or buying. S&P 500 futures should trade around their "fair value," which is a premium over the S&P 500 index number based on interest rates and dividend flow. If buying or selling moves the price of the futures too far above or below the fair value, arbitragers can take advantage of this mispricing. But if the price of the futures is pushed below its fair value, it is because people are selling futures instead of stocks. If the futures market weren't there, these people would be selling stocks. It's almost like a simplified Rube Goldberg diagram. Instead of selling stocks directly, you sell futures which causes someone else to sell stocks because you have made them overvalued compared to futures. Once again, computers are blamed without much real knowledge of the process involved. The conventional wisdom is that program trading is completely automatic, without human intervention. Yet it is the human trader who is in control of the two transactions involved — one talking to a guy in a futures pit shouting and making strange hand signals, and another typing a few keystrokes at a keyboard. Dan Blumenthal firstname.lastname@example.org
In light of recent Wall Street instabilities and previous discussion in RISKS about computer voting fraud, some history might be in order. Thomas Alva Edison's first commercial invention was an electric voting booth. The votes were electrically and instantly tabulated. Unfortunately, American politics at the time wasn't interested in either fast or accurate vote-counting. The machine was a commercial failure. Edison vowed then and there never to develop a machine that there wasn't a ready market for. His next invention was the ticker-tape. In light of the ensuing century, perhaps we would have been better off if the response to his machines were reversed :-). Brent Laminack (gatech!itm!brent)
WTOP, the local all-news radio station in Washington D.C., reported the following on 28 October 1987. Unfortunately, I was riding along in my car when I heard this, so it isn't verbatim. 3 years ago, at a SAC missile base in the midwest, a Minuteman III missile malfunctioned. The missile was carrying three nuclear warheads, and went into launch mode. Technicians couldn't understand what was going on, since there was no alert occurring at the time. Their immediate reaction was to take a large armored personel carrier (APC) and park it on the missile silo. Assuming the silo doors opened, the APC would fall into the silo, and hopefully stop the missile. Missiles apparently do not arm their payloads until they are off the ground. The report went on to say that the missile did not launch, and that the fault was tracked to a problem in the guidance system. The incident was never reported to the Strategic Air Command, local officials, or apparently anyone outside the missile base. According to this, the only way we can stop a malfunctioning missile is to drop a big 'rock' on it. John McMahon, FastEddy@Dftnic.Gsfc.Nasa.Gov (Internet), FastEddy@Iafbit(Bitnet)
Please report problems with the web pages to the maintainer