Computers in Space Examined

Wil Harris writes "There's an article about the computers used in space missions over at bit-tech this morning. It covers the processor types and speeds, why space stations are less powerful than the laptops that astronauts take up with them and why tape storage is still de rigeur. An interesting and concise couple o' pages."

Comments

Reminds me of a story I was told

[Raul654]

Dave Mills (inventor of NTP) told me that on the last Columbia shuttle mission, they were running some experiments with NTP in space. And, thankfully, they transmitted all their data before landing. But apparently, they were so overworked, they didn't have time to calibrate the machine properly, so sadly, the data is useless.

Solid state recorder on board the Hubble

[helioquake]

Just like Cassini, the Hubble also has on-board solid state recorder (installed during one of the servicing missions), which replaced an old tape recorder. This has been really a nice addition as we can store more data into the solid state device while collecting data bits and dump them when the downlink becomes available. It really helps increase the efficiency of the satellite (and that's a big thing for science mission).

[Note that I've simplified the scheme alot here.]

Though several sections of the device have been damaged by radiation, or something, I hear. So even these things aren't too resilent to the harsh space environment, yet. Something you future engineers should think about as a project.

astronauts use winamp in space

[Anonymous Coward]

see http://spaceflight.nasa.gov/gallery/images/station /crew-1/hires/iss01e5127.jpg

GPC vs. embedded

[fermion]

I think what most people miss is that the devices in space craft, hell even most cars, and not general purpose computers(GPC), have no need to be GPCs, and should not be GPCs. A GPC, quite frankly, is jack of all trades. Designed to do everything adequately, nothing well, and is a single point of failure. It might be over-designed to do a few things well, but then it costs more.

An embedded machine, OTOH, is designed to do one, or a very small range of things, very well, very reliably, and very efficiently. I have had the fortune of working on two space based projects. In the first we used a single board Z80 based space hardened 'computer' to control a simple set of devices. It stored the ASM code in an EEPROM. It was more complex than we needed, as it was a standard issue unit, but much simpler than the Apple ][ we used as the GPC.

On the second project, 10 years later, we were not using incredible different machines on the satellite, though the GPC was now a Wintel machines with 100X the memory and speed. But when your main concern is that things just have to work, processor speed and OS wars have little meaning.

So these stories about how underpowered and behind the times embedded systems are just annoys me. It is just like continuous burns on SciFi shows(kudos to Babylon 5). Perhaps meaningless power is important to the ignorant masses, but we on /. are supposed to know better. I was using a tape drive until at least '87, just because It Worked.

Re:Tape?!?!

[bluGill]

They are not useing the junk tape drives that you were using, but quality stuff. Mainframes have always put most of their data on tapes drives, and they rarely have problems.

Course a mainframe tape drive can cost $30,000 each, (not counting the robots that load them) so you can see why home users don't get that quality.

Determinism

[Detritus]

One important point that the author missed was determinism. Many of these computers are used in hard real-time applications. If the tasks don't meet their deadlines, the system has failed. This requires predictable timing. That means cache is a liability, as are many of the advanced features of modern processors. You need to be able to sit down with a program listing and count how many CPU cycles it takes to execute a segment of code. If you have a 10 ms time slot to do a task, you have to be able to prove that the code can run in less than 10 ms. If you poll a set of sensors every 100 ms, you want the timing to be identical every time the code runs, to eliminate timing jitter in the measurements.

The controller for the SSME (Space Shuttle Main Engine) uses a pair of 68000 processors. It is a very critical system. If something starts to go wrong with the engine, it has to detect the problem and shut the engine down before it progresses to a catastrophic failure. It uses two redundant processors for reliability. Each engine has its own controller.

Old microprocessors like the 80386 and the 68000 were the last commercial processors before cache, pipelines and other trickery made timing analysis difficult or impossible. Some people have used DSPs for controllers because they still offer predictable timing.

Re:Article's missing/wrong on a few points

[helioquake]

There was a slashdot article on this a long, long ago...this might be it (not original, however):

They Write The Right Stuff

It's a must read for programmers at mission critical stuff.