Columbia crypto box

[cryptography at ka9q.net]

> 	As an aside, I've been seeing a *lot* of criticism in the popular
> press about the alleged antiquity of the STS computers; all asking why
> NASA isn't using the latest technology in its shuttles.  Folks, I ask you
> to take the latest bleeding-edge technology and subject it to significant
> G-forces followed by introduction into a microgravity environ, have it
> bounced around during re-entry and then _guarantee_ that it won't
> experience a critical failure at the worst possible moment.  That
> floating-point error in your Pentium may seem catastrophic when you're
> working in AutoCAD, but just have a look at the French Arianne rocket if
> you want to see what such an error does to a spacecraft in flight.

It's interesting that you would cite that particular failure, because
it was ultimately caused by the use of obsolete computer hardware.

The software writers were well aware that certain floating point
variables might cause an exception when converted to fixed point. They
even added range checks to many of these conversions. But they
couldn't range check *every* variable before conversion because they
didn't have enough CPU cycles; a design rule required that peak CPU
utilization remain below 70%. So they had to pick and choose which
conversions would be range checked. They deliberately left "horizontal
velocity" unchecked because it wasn't possible (on Ariane 4) for that
variable to increase quickly enough to cause a problem if the launcher
flew normally. And if it wasn't flying normally, who cared? But Ariane
5 was designed to gain horizontal velocity much faster than Ariane
4...

Had they used a newer, faster CPU they would have been able to range
check *every* variable before conversion, or use a more modern
architecture that wouldn't cause a fatal exception on an out-of-range
conversion. But hardware conservatism prevailed, with disastrous
results.

Phil


---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo at wasabisystems.com