Overall the space shuttle software is definitely more complicated than confirming a small prime, but on a line by line basis I don’t know. The space shuttle is more of a special case example; not something you compare against.
I’d use the industry average rate until you showed me you could hit 0.1 errors/kloc for a few years. For example, Microsoft hits 10-20 defects/kloc and they put significant effort into testing.
The most likely reason that your bug rate on these programs would be anomalously low is the fact that they’re small. Instead of writing one hundred thousand line program, you’re writing 10000 ten line programs. The complexities won’t compound as much.
The probability of mistake in a program is not the same as the probability of being wrong about 1159 being a composite. At high reliability level, you factor the number, then you multiply up the factors, and check that as well.
Overall the space shuttle software is definitely more complicated than confirming a small prime, but on a line by line basis I don’t know. The space shuttle is more of a special case example; not something you compare against.
I’d use the industry average rate until you showed me you could hit 0.1 errors/kloc for a few years. For example, Microsoft hits 10-20 defects/kloc and they put significant effort into testing.
The most likely reason that your bug rate on these programs would be anomalously low is the fact that they’re small. Instead of writing one hundred thousand line program, you’re writing 10000 ten line programs. The complexities won’t compound as much.
The probability of mistake in a program is not the same as the probability of being wrong about 1159 being a composite. At high reliability level, you factor the number, then you multiply up the factors, and check that as well.