Embedded Muse 75 Copyright 2002 TGG September 25, 2002
You may redistribute this newsletter for noncommercial purposes. For commercial use contact email@example.com.
EDITOR: Jack Ganssle, firstname.lastname@example.org
- Around the Net
- Book, and a Comment on Disaster
- Thought for the Week
- About The Embedded Muse
Around the Net
Good friend Bill Gatliff is offering Embedded GNU and Embedded Linux classes. Visit his site at http://billgatliff.com. He’s always a popular speaker at Embedded Systems Conferences.
Check out http://ee.cleversoul.com, The EE Compendium. A very nice collection of reviews of EE books and links to industry sites. Also check out sublink http://ee.cleversoul.com/software.html, which has links to bug tracking and version control utilities. A lot of folks ask me about these.
Book, and a Comment on Disaster
I just finished “Inviting Disaster” by James Chiles (HarperBusiness; ISBN: 0066620813; September 2001). A very interesting book, well worth reading. Chiles describes a number of technological disasters. He delves into the root causes of each. Some of the events were averted before anyone was hurt; others resulted in hundreds and thousands of deaths.
I'll present my Better Firmware Faster seminar in Melbourne and Perth, Australia February 20 and 26th. All are invited. More info here. The early registration discount ends January 20.
A lot of failures were at least partly caused by instruments that lied to the operators. In two of those cases (Apollo 13 and Three Mile Island) temperature sensors read max values (at TMI 280 degrees, though the input was closer to 1000) simply because that’s as far as the programmers allowed the instruments to read. Any higher value was thought to be impossible. This is a tough problem for us. Specs are notoriously incomplete; it’s likely that in these cases developers made decisions that seemed reasonable at the time, decisions that worked very well until an unlikely series of events unfolded.
Chiles doesn’t cover the Therac-25 (see http://courses.cs.vt.edu/~cs3604/lib/Therac_25/Therac_1.html) which killed a few patients. We embedded people know of it. But that was 1985, software was much more casually written then. Today we don’t tolerate such error-prone methods.
Maybe we do. Check out http://www.iaea.org/worldatom/Press/P_release/2001/panam_adv_info2.shtml. Turns out another therapeutic instrument, meant for much the same application as the Therac-25, recently killed 9 people in Mexico; as of the initial report date (May 2001), another 15 or so were expected to die. It was a complex failure due in part to operator errors. But the machine lied to the users, telling them the programmed dosage was safe. It wasn’t.
Civil engineers study old bridge failures. Aircraft designers have a wealth of information from plane crashes. We, too, cannot afford to thwart disaster by learning solely from our own experiences.
Thought for the Week
This issue’s Thought for the Week comes from responses to the last Muse’s discussion of the IEC’s definitions of prefixes for powers of two. To reiterate, the IEC has defined the following prefixes:
2**10 kibi (symbol Ki) (one kibi is 1024)
2**20 mebi (symbol Mi) (one mebi is 1 048 576)
2**30 gibi (symbol Gi) (one gibi is 1 073 741 824)
Tom of Bunyon commented: “For some people, this is byteing off more than we can queue...”
Dave Kellogg made the interesting observation that 1000 terabytes is a petafile…
I like Tim Kramer’s thought: "There are 10 kinds of people in the world: those who get binary and those who don't." (Think about it).
Lance Kasari brought up a fascinating problem with the sytem. People will try and mix base 10 and base 2. So 1000 mebibytes does not equal one gibibyte (which is properly 1024 mebibytes). He figures that 1 gibibyte is really the same as a kibimebibyte (1024 mebibytes)… but how do we keep people from giggling when they say this?