The systems are time-tested. How many accidental launches have happened so far? Zero. In my opinion that's the most important metric of these systems, as long as they are still credible threats otherwise.
Also, what could a better system improve? Slightly faster launches? Slightly more certain launches? I argue that those are not important -- the enemy won't fear the missiles more if they run off USB. But you might increase the chance of a false launch, and certainly incur large testing costs.
From a military point of view, the huge advantage of floppy disks is that they're dumb storage. The modern replacements such as USB storage or SD cards are microprocessor-based systems. How are you going to verify that the SD cards you're buying don't have malicious firmware?
Adding to that, 8 inch floppies have a certain quality to them that no younger medium has ever recreated: try hiding one on your body, try sneaking in an illicit reader to make "personal backups"... Security through unwieldyness goes a long way, when your biggest threats are of the social kind.
Uhmmmmm, read Eric Schlosser's "Command and Control". The US nuclear missiles had no protection against accidental lunch, you didn't even have to put in a code or a key, and the US government has said, sincerely, that "the US nuclear stockpile is completely safe, based on their observation that there was no accidental lunches yet".
If you have raw electrical cables sticking out of your wall, they are not "safe" just because you haven't electrocuted yourself yet.
"The systems are time-tested. How many accidental launches have happened so far? Zero. "
Hell, if that's your metric just unplug the thing. Exactly zero accidental launches, and you'll only have made the wrong decision once if anything goes wrong...
You should buy this talisman of mine that wards away tigers. You know it works since I've had it my entire life and I've yet to be attacked by a tiger!
Completely agree with this. I actually think vims undo tree might be the most useful feature in vim for me. You can basically use it as version control lite with automatic commits.
Norton Commander was a great concept, but what really perfected that concept was Dos Navigator... It had so many useful features... If you didn't use it in the 90's, you missed out on a legend! It was open-sourced after DOS became history: http://www.dnosp.com/.
And there was also Volkov Commander, which was about feature-equal with NC, but was written by hand in assembly. Tiny resource usage and lightning speed.
I'd never heard of "drag catastrophe" before either, but I think you're right that he's referring to the onset of turbulent flow that reduces drag on a bluff body.
That's the sudden drop at the right in all of these graphs: www.google.ro/search?q=drag+reynolds+number (same as his own Figure 9, really).
There are a lot of surprising nonlinearities in those graphs. I wish there was a more detailed article/paper somewhere.
I can't understand how people trust (or even care about) server-provided encryption. There's no guarantee they or their friends can't open it on their servers. Unless you use your own solution, their encryption only provides protection in transit, against unaffiliated third parties.
Link encryption is good, and it's helpful to ensure the only attack vector is the service provider itself, not sub-contractors or someone being stupid and throwing away drives.
That said, Dropbox has done a pretty good job of lying about how much security they offered users in the past. But someone providing a service like this with a reasonable security policy which is openly communicated to the users is still better than no security policy.
When it comes to zero-knowledge encryption, it is hard to know how their implementation holds up without external audits. This is the same for other encryption applications (google Truecrypt audit).
Also, what could a better system improve? Slightly faster launches? Slightly more certain launches? I argue that those are not important -- the enemy won't fear the missiles more if they run off USB. But you might increase the chance of a false launch, and certainly incur large testing costs.