During the course of trying to figure out why AA587's tail came off, the US NTSB is looking at an incident to another Airbus A300-600 aircraft in 1997, also in service with American Airlines. AA Flight 903, en route from Boston to Miami on 12 May 1997, experienced an upset after descending to 16,000ft in preparation for landing. The NTSB determined that the aircraft stalled, and entered pitch, yaw and roll manoeuvres described as "oscillations" which continued for 34 seconds before recovery. The aircraft lost 3,000ft altitude during the event. The NTSB also determined that the crew took "improper remedial action" after the aircraft was allowed to stall. One person was seriously injured. Aviation Week (4 Mar 2002, pp52-3) says that investigators learned that flight data displayed on the Electronic Flight Information System (EFIS) screens disappeared for 2-3 sec. during the upset while the avionics reset. "Data were deplaced by white diagonal slash marks across the screens." That means that the crew lost essential flight data: attitude, airspeed, rate of descent, altitude, etc. This data would normally be essential to proper recovery from an "unusual attitude", particularly at night and in clouds. (The AvWeek article does not state the time or weather conditions.) As a consequence, the NTSB issued safety recommendations A-98-3, through -5. A-98-3 asked the FAA to require modification of the Symbol Generator Unit software so that "unreliable data reset of the [EFIS] will not occur during an upset". The SGU renders the flight data on the EFIS screens from sensor and other input. The NTSB says it "learned that the threshold for triggering an auto reset can be reached during an inflight upset. For example, if the roll angle rate of change is more than 40 deg. per sec., a reset will occur." According to the Flight Data Recorder, this limit was reached during the upset. Peter B. Ladkin, University of Bielefeld, Germany http://www.rvs.uni-bielefeld.de
On 20 Mar 2001, a Lufthansa Airbus A320 destined for Paris came within two feet and a few seconds of crashing on takeoff from Frankfurt. The captain (in the left seat) was Pilot Flying (PF) on takeoff. Shortly after leaving the ground, the aircraft encountered a little turbulence and the left wing moved down. PF corrected, applying "right stick", but the aircraft responded by rolling further left. Left bank reached 21 deg. and the left wingtip came within 1.5 feet of the ground. The first officer, Pilot Not Flying (PNF), realised what the problem was, switched sole control to his stick (normally, the control inputs from both sticks are averaged) and recovered the aircraft. The crew climbed the aircraft to 12,000ft, performed some handling checks to confirm that the captain's stick was operating in the reverse sense in roll, and landed back at Frankfurt. Control reversals of this sort are known with conventional aircraft, in particular, small aircraft with cables between cockpit controls and aerodynamic control surfaces. Cables are reconnected in reverse during maintenance. It pays to check the controls thoroughly and carefully after maintenance (in fact, one is required to check them before every flight). Not everyone can recover - has recovered - either themselves or the aircraft from the surprise of suddenly having to deal with controls operating in reverse sense upon takeoff. But the A320 is fly-by-wire, not fly-by-cable. It turns out that the captain's controller had indeed been reverse-connected (electrically) for roll commands, during maintenance. And the first officer's was correct. With cable control, if one is wrong, then both are wrong. In either case, this should be noticed on the control check before takeoff. The incident raises three main questions: 1. How did the misconnection happen? 2. Why did maintenance not discover it on the return-to-service control check (after controls have been worked on, such a check is mandatory)? 3. Why did the pilots not discover it on the pre-takeoff control check? Sidestick roll commands are input into five computers: two Elevator and Aileron Computers (ELACs) and three Spoiler and Elevator Computers (SECs). Each of these boxes is "dual channel", with two processors, one hot and the other shadowing the first. Each ELAC contains a pair of MC68000 series processors with dissimilar software; each SEC a pair of Intel 80186's with dissimilar software. For roll, the ELACS control the ailerons, and the SECs the spoilers, on the wings, through three different hydraulic systems. Roll is achieved through use of the four outer spoilers (of five, per side) and ailerons (two adjacent, outboard of the spoilers, per side), and actuation of these surfaces is distributed amongst the hydraulic systems to achieve redundancy amongst the hydraulics. Control command redundancy is achieved by distributing the control surfaces amongst the computers, through different hydraulic systems. One can even lose both ELACs and roll is then controlled by the SECs, and vice versa. A diagram and explanation can be found on pp. 133-4 of Cary R. Spitzer, Digital Avionics Systems: Principles and Practice, Second edition, McGraw-Hill, 1993. The incident aircraft had returned from maintenance. According to the report by David Evans in Air Safety Week (ASW), 4 Jun 2001, maintenance personnel had found a damaged pin on one segment of the four connector segments on the "rack side" of one of the ELACs. Each connector segment has 140 pins. ASW says that a complete rewiring "upstream" of the connector pins was performed. Apparently the polarity was reversed on four wires in one segment: two for roll control inputs and two for the associated control channel outputs. Although it is physically impossible to mismate connectors, it is mooted that there are some differences in color-coding of the wiring between different aircraft models and that this may have played a role. But this story conflicts prima facie with the details of the architecture and the incident history. The sidestick input goes into five computers, which amongst other things vote somehow on consistency. If only one ELAC received reversed control commands through reversed wiring, it would have conflicted with the other and they would have been taken off-line, leaving the SECs with (correct) control. Generalising, it follows that a majority of the five ELAC and SEC computers were operating with reversed control from the left sidestick during the incident. The reversed connection must have been effected upstream of the point or points at which the one control signal from the sidestick is demultiplexed into the five signals for the five computers. If a connector and the associated wiring to one ELAC was faulty, I do not see why that should require a rewiring upstream of any demultiplexor. I do not see how such a rewiring can have affected signals to all five (or even a majority of all five) computers. (The other possibility is that the signal from the sidestick is not demultiplexed; that each computer receives a separate signal direct from the sidestick. But that would require at least ten pins on the sidestick connector - two for each computer - and only four wires were reported misconnected; even those were in a two-plus-two arrangement. So that could not have affected the senses of the other six.) It has been confirmed that, after maintenance, the technician performed control checks only on the right-seat sidestick, not the left-seat stick. It makes no sense to me why, after performing maintenance on the left-seat sidestick, anybody (let alone a Lufthansa technician) would then perform checks only on the *right-seat stick*. If you have just repaired a flat tire on your car, you don't inflate the tire on the other side! That suggests some cognitive confusion, namely that the rewiring did not take place right up to the sidestick, but up to an intermediate connector that was imagined to be common to both sticks. But, as I have just noted, I just cannot reconcile that supposition with the avionics architecture and the rest of the story. The preflight checks require both pilots to perform a control check. When the check is performed, a schematic of the control surfaces (the "Flight Control Page") appears on the screen of the Electronic Centralized Aircraft Monitor (ECAM), and the actual control outputs are shown. The ailerons and spoiler actions would have been shown reversed. Upon performing a control check in this situation, it is just conceivable that one might not notice that the ailerons are operating in reverse sense, but the spoilers are asymmetric: one side goes up and the other doesn't move. I cannot imagine putting in left stick and then just not noticing that the *right spoiler bank* activated instead of the left bank, or vice versa. So there is another puzzle. And if some of the computers were computing differently-sensed control commands from others, one would have expected that an annunciation on the ECAM to that effect would have followed, and that the annunciation would have been noticed by both pilots. In the ensuing ten months, I have obtained no information which sheds further light on this incident. There are two kinds of issues here. One kind admits possible explanations: maybe the humans failed in a big way on their post-maintenance and pre-flight checks somehow. The other kind does not admit any explanation: I see no way to reconcile the supposed details of the misconnection in this incident with the architecture and operation of the ELAC and SEC avionics. The story must be different from that which has been told, but I do not know how. Peter B. Ladkin University of Bielefeld, Germany http://www.rvs.uni-bielefeld.de
>From AVflash 8.11b. <http://www.avweb.com> Later this week, representatives from NASA, the U.S. Navy, New Mexico State University and the industry will descend upon Las Cruces, N.M., to show that remotely controlled aircraft can operate safely in the National Airspace System. The team will fly up to three aircraft on collision courses to test onboard sensor technology designed to detect and avoid potential threats. The Proteus aircraft, built by Scaled Composites in Mojave, Calif., will take part, fitted with a Skywatch HP traffic advisory system, a radio-based device that detects other aircraft. Proteus will also carry an Engineering 2000 infrared sensor, and an Amphitech radar "non-cooperative" sensor -- devices that don't require signals or transmissions from any other source -- to detect the presence and course of other aircraft. [Gives me a nightmarish vision of a cloud of little unmanned aircraft all heading for the same place, trying to avoid each other, and the regular, piloted airliner flying through the cloud and scattering it in all directions - pretty scary. EK]
My daughter was finally called in to aid with a help desk call from a Japanese visitor to Perth Australia. He was setting up his Mac to connect to a local ISP with a package the ISP had supplied. The visitor was entering into the dialog boxes the usual english texts "firstname.lastname@example.org etc" but it would not connect. Many phone calls to the support desk were futile. Then my clever bilingual daughter who uses a Japanese OS Mac herself finally realised the Mac user was probably in some Japanese text mode? when entering the "English" looking characters into the boxes. When she got him to change to a pure English text mode and entered it all again it worked! I guess that when the connect strings went out instead of sending 4 ASCII bytes for "fred" it was sending out 4 unicode 16 bit codes, or 8 bytes. The OS was storing the visibly "English" text somewhere as unicode and the modem string handler was just sending out a (8 byte) null terminated string. Maybe the problem was that the ISP's application was not Unicode aware, but definitely a really nice case of WYSINWYG. What You See Is NOT What You Get.
Lars Vilks, state secretary of Ladonia, reports that more than 3,000 Pakistanis recently applied for Ladonian citizenship over the Internet. According to an 11 Mar 2002 CNN.com SCI-TECH report, Ladonia already has 6,000 registered "citizens". According to the official national Web site (http://www.aim.se/ladonia), Ladonia is one square kilometer in size, between Sweden and Denmark, having become a free nation on 2 Jun 1996. Its capitol is Wotan City, with one main wood building (Nimis) and a nearby town (Arx) with a stone and concrete construction. The national anthem is performed by throwing a stone into water. Having been swamped with applicants, its Web site was temporarily shut down -- because applicants had mistakenly expected jobs and housing. The on-line citizenship application has now been amended, with an appropriate disclaimer. Common citizenship is free; nobility costs $12, and you can choose your own title. In actuality, Ladonia exists only on the Internet and in Vilks' imagination. According to the CNN report, Vilks established Ladonia on 2 Jun 1996, protesting Swedish authorities who had attempted to remove his two large abstract art works (Nimis and Arx?) from Skaane in southern Sweden. Please get your April Fool's Day items in early. This is not one of them. Everything above is true, except possibly my question-marked supposition about the identity of the disputed art works.
I received the message shown below from Worldnet. Note the quoted first line. I absolutely refuse to use any mail reader that "supports" html. Howver, everything after <META- in the message below appears as an html box in Agent. In other words, unless your mail reader supports html you will never see the message telling you what to do! Perhaps AT&T should have tested this on someone who actually uses software that doesn't support html. - Tony Lima On Wed, 6 Mar 2002 20:11:03 -0500 (EST), attwns-announcement-system <CustomerNotifications@worldnet.att.net> wrote: >To: AT&T WorldNet Customers <WorldNetCustomers@att.net> >Subject: Your March Newsletter Is Here! >From: attwns-announcement-system <CustomerNotifications@worldnet.att.net> >Date: Wed, 6 Mar 2002 20:11:03 -0500 (EST) > ><META HTTP-EQUIV="Content-Type" CONTENT="text/html;charset=iso-8859-1"><!-- If your e-mail tool doesn't support html, please see the on-line version at http://www.att.net/perks/newsletter/ -->
>From the BBC News Website: http://news.bbc.co.uk/hi/english/sci/tech/newsid_1860000/1860241.stm The article notes that an antenna for eavesdropping on Wireless Networks can be created out of an old Pringles tube. A link from this article goes to a page describing how such an antenna can me made (this article by Rob Flickenger). It is written in a partly humourous vein, and is well worth a glance: http://www.oreillynet.com/cs/weblog/view/wlg/448 This is of interest to RISKS not so much for the security/privacy risks inherent in a wireless network (see RISKS Passim), but for how easy and cheap it is to create the tools for the job. The ability to create antenna "out of anything" is nothing new: I remember my Antenna Theory lecturer describing how a satellite dish could be created out of an old metal dustbin lid. That was about 20 years ago.
A few years ago, my group at Sandia National Labs was building a new operating system for the Intel Paragon massively parallel super computers. These machines had an amazing high-speed message passing interconnect, but it was flakey while we were rebuilding the network driver. The nodes were nearly "embedded", with only the mesh interconnect and a few LEDs on the front panel. They had no other IO, not even a serial port. There was an even more unreliable channel used for downloading kernels, but it failed as often as the mesh driver. To help get crash data off the nodes, we wrote an ISR that would translate printk() calls into 2400 N81 pulses of one of the front panel LEDs. We used a 4k ring buffer, so it would continuously repeat until the machine was reset. A light pen was duct taped over this LED and hooked to a nearby SPARCstation's serial port. We used this for debugging for several years and through many versions of the operating system. The software eventually went into "production" use on the larger machine. We occasionally saw nodes that had crashed on the big system blinking out their panic messages. Since it was installed in a secure computing facility, I was always concerned that this might be used as a covert channel for extracting data (slowly!) from the classified compute runs. > ... and it was difficult to hold the detector in the exact alignment We had no such problems. I seem to recall that this was working within a weekend of realising that we could do it. Most of the time was tracking down the light pen... Recently a similar subject was discussed on alt.folklore.computers. <email@example.com> started the thread on "Morse code diagnostics". Trammell.Hudson@celera.com http://www.swcp.com/~hudson/ W 1-240-453-3317
LEDs intended as activity lights for human observation are engineered to give bright and easily perceived output when signals are detected. This requires flashing at a few Hertz max, despite the signals being anywhere between 2400 bps and 100+ MHz (depending on the system). The solution is to use a monostable or other pulse-stretching device in the driver. This destroys any direct relationship between LED brilliance and the data. Direct drive of the LED with the data is not normally used - short data bursts just are not seen at all and continuous data usually gives the half brightness effect described by Nick Simicich in Risks-21.95. I am prepared to believe that LEDs indicating modem *status signals* such as CTS (Clear To Send) *might* be directly driven by the signals, and thus *might* be viewed at a distance - but I am not prepared to believe that the *data* is exposed in this way. [Note that many important advances in knowledge have been immediately preceded by an expert saying "this cannot be done". I am therefore making my contribution to this topic by saying "this cannot be done" (!)]
> You are talking about the rise of the Apostrophe Virus [...] Ah, but folks who read these electronic documents on anything but Windows see not apostrophes, but question-marks or little hollow rectangles, courtesy of the exceedingly ill-named "smart quotes". That is, they do unless the author has been silly enough to include the apostrophes yet wise enough to run the text in question through John Walker's "Demoronizer". This appears to me an unlikely convergence. I have even seen cases where the offending apostrophes were completely eliminated, which would be a good thing (tm, Martha Stewart), if not for the fact that the correct apostrophes were also eliminated. > ..., and spellcheckers are assumed to handle all grammar issues. Only Hogwarts students really need spellcheckers, IMHO :-) (although perhaps I am mistaken, given the Sorceror's Apprentice nature of the average Spelling Checker)
I read with interest Gene Spafford's tale regarding his friend's addressing within the 69 Class A netblock. I just did a quick rifling both the ARIN database and the IANA IPv4 Addressing Space, because I too have my routers and firewalls similarly configured. It would seem that both ARIN AND IANA are just as unaware of the assignment of the 69/8 netblock as Gene was. Both list the 69 through 79 Class A's as "Reserved." I think this merits further investigation.
It was a typo in my posting. It was the 67.* block, not 69.*
And the risks are apparent. ;) (Sorry, couldn't resist.) Looks like the 68/8 was picked up a month after the 67/8, too. Time for me to update my lists...
[Gene Spafford writes about a problem with a router's configuration, and how he set up a periodic e-mail reminder.] I think your RISKS example points out one of the most difficult issues with system administration: documentation. Mr. Spafford has slapped a temporary bandage on the problem by sending himself an e-mail every 6 months, but he's just traded one risk for another. What happens if his 'at job' (or whatever) fails? (*) The only time he'll realize that it has failed is when he didn't need it (he remembered anyway). And what about the ten thousand other bits of knowledge that he has in his head that helps keep his network running smoothly? Setting up at jobs for all those bits of information isn't practical. And then, what happens when he's on vacation, and doesn't want to be disturbed? Of if he's been hit by a bus? None of this is meant as a personal attack on Mr. Spafford. Having relied on his advice (through the printed page), I have only the highest respect for him and his efforts at computer security. It's just that his response to the situation is typical of a system administrator. The better system administrators tend to have very good memory, which is both a blessing and a curse. It helps tremendously to not have to constantly refer back to documentation to get something done. Unfortunately, the one-off problems (that will _never_ occur again :-) ) don't tend to get documented. The busy sysadmin feels he has better things to do than spend 15 minutes writing up a problem and solution. The real RISK with all this attitude is that as our lives become more complex (and dependent on technology) it will take longer and longer to diagnose and fix problems. And worse yet, we'll be fixing the same problems over and over again. The only counter to this RISK is documentation. Every organization ought to have (IMHO), a central jumping-off point for documenation. There has to be one place for people to start looking for answers, if those answers exist at all. Documentation is great, as long as it's relevant. As it gets out of date, it's less and less likely that people will use it. So then they'll spend (or waste depending on the point of view) time diagnosing and solving the same problems again. And in the mean time, service will suffer. The best tool I've seen for up-to-the-minute documentation is a Wiki. If you can encourage the 'wiki attitude' in your organization, you'll end up with a really good, and relevant reference, which will go a long way towards improving the work. James Graves (*) Aside from software bugs, the classic reason why 'at' jobs fail is because people forget about them when they are moving to new machines. The ones that run every day are remembered very quickly if absent. But those that only run every six months...
Subject: Re: Sorry, that number is now in service Actually, I documented it in 3 places: * in the configuration file, near where the rules are * in a readme file in the same directory, where I keep notes for whoever maintains the file * by posting to Risks so we build up community memory! :-) Unfortunately, in first posting in RISKS-21.92, I misidentified the networks involved. :-)
Michael (Streaky) Bacon (RISKS-21.95) complained about the message on an e-mail from the BBC (British Broadcasting Corporation) stating that e-mail may be monitored and that further use indicated acceptance of that monitoring. That sort of statement will be seen more widely on messages coming from the UK because it is part of a requirement of an Act of Parliament called the "Regulation of Investigatory Powers Act". This is combined with a Statutory instrument called "The Telecommunications (Lawful Business Practice Regulations) 2000" . The requirement is that monitoring may only be carried out "if the controller of the telecommunications system on which they are effected has made all reasonable efforts to inform potential users that interceptions may be made". How can one inform e-mail users other than a message on an e-mail? Many of us are still pondering how we warn people who contact us for the first time..... John Hitches
Please report problems with the web pages to the maintainer