I mean I don't expect my desktop tower PC to run flawlessly 24/7 for over a hundred years
Not your desktop, and with somewhat more maintenance, but I've looked after systems 15+ years old that are still going strong. They are still going strong as it happens, but are no used directly by me. Good design, multiple redundancy and high quality you can do that easily. 100 years with regular maintenance would be easy by itself, the problem these days is finding the replacement hardware for Alpha's and Vaxen. This isn't an isolated example, my organisation still has machines from the early 80's still in service and still useful.
As for the space travel we might not be able to test that aspect yet
Unless there was a lot of unsheilded RF these machines should be able to stand up to that just as well.
Of course these machines originally cost millions of dollars, require people constantly maintaining them and are the size of a large two door fridge for something that most modern PDA's would be able to out-process. But for reliability there is no real match.
I am sure main frames are a creature of the past for most purposes.
You'd be surprised. A lot are creatures from the past, but a
SUN20K is a nice little machine all on its own.
As is a
IBM Z990 which is almost a dTon in size. 5 m^3 minimum, but with ventillation, service access and enclosure is 11 m^3.
You can go out and buy either tommorow. Get them delivered in time for late this year. And then scrounge up the 2-10 million $US to pay for them.
Of course so far the track record on the newer machines (the SUN, I have no direct contact with big iron IBM) is pitiful compared to the older machines.
Now from a service point of view you can wear your redundancy inside or out. The farm configuration (using a large number of expendable machines all to support a single service) takes up a similar ammount of space to the above big iron and can for most purposes be just as good. The failure of machines is built into a farm, you need to keep pouring new machines into the service to maintain its quality. What this leverages is the extreme cost advantage of consumer grade computer gear.
For example the primary system I look after these days runs on PC's. A direct statement, and at first implying that PC's are a good thing. They are. The service runs on 13 high quality low grade intel based computers. Because the service can run with some lost data, and the individual nodes can act independantly it works well with a farm configuration, so that is what it is in. There is also a multiple located disk array accessed via high speed dedicated fibre, as well as a shared backup facility. All this adds up to a dTon+ of gear and needs to be constantly fed sacrafices to maintain stability, not to mention new tapes and disks as the old ones wear out. This runs a glorified automated information counter (ie. a website) and does nothing else.