Brian Gyles made some interesting comments:

>I remember reading they only ran the program for some thing like 16 days for
>testing. when testing you should get as near as you can to the real thing
>stoping the test short of the expected life cycle is not a good idea.

I suspect (don't know) that this was a memory leak. What got JPL into
trouble was not the time duration that MER-A ran but the amount of data
they tried to store in Flash memory. I've had the experience with
memory leak bugs of creating the error but not having it bite me until
months after the error was made. JPL should have torture tested their memory storage software by loading in gigabytes of test patterns. However the bug might have been something nasty where test patterns didn't trigger it but Mars photographs did, i.e. data garbage accidently matching a machine language op code.

>they should be using 3 processors/programming languages tied to a vote
>system if 2 come forward with the same response do it. they use to do it for
>mission critical systems. but even this approach has problems as if the
>programing teams are one of the same or trained in the same manner then
>the structure and tecqueues could contain the same inherted problems.

I was surprised when I read that MER used only one processor (a RAD6000) with no backup processor. However the electrical engineers at JPL are not stupid. There is most likely a method behind their madness. I suspect someone calculated the probability of failure for a system using different isolating tri-state gates protecting the redundant processor data buses versus the probability of a system with one super reliable processor. It's quite possible they found keeping it simple with one processor lead to a more reliable system (it would certainly burn less power). This sort of design process is not intuitive.

>one thing that shows different design / test cultures is that ibm mini
>computers have a lower lock up while being used rate than wintel based
>systems.

That's damning IBM with faint praise. Wintel based systems are excrement (at least prior to Windows 2000). I don't like Microsoft software but I've found Windows 2000 to almost border on acceptable. I only use Windows 2000 to run MS-Word and play card games. I never access the Internet from a Microsoft operating system (they're almost completely unsecure). When I'm doing serious work or accessing the Internet, I always use Linux. At work we have Beowulf clusters with dozens of Intel Xeon processors that run for days on Linux and never have a hiccup (I prefer the AMD processors because of their better bang-per-buck). At Lawrence Livermore they run bomb codes on enormous Beowulf clusters using thousands of processors operating under Linux. Supposably none of the Microsoft operating systems have the reliability to function in that environment.

Gary