News for hackers. Stuff that matters.
Re:Darpa Seeks to Create Software Systems That Could Last 10... by dragonwriter on Wednesday April 29, 2015 @06:04PM (#9462145)
No, for much the same reason that in, say, 1945 they weren't using any airframes that had been in service for 50+ years.

OTOH, B-52 production deliveries were 1954-1963, there are B-52s in service now, and the USAF currently plans to keep them in service through 2045.

Re:Darpa Seeks to Create Software Systems That Could Last 10... by greggyb on Wednesday April 29, 2015 @07:10PM (#9462427)
If you replace every piece of a B52 with a new piece, is it still the same plane?

Especially if the new pieces are no the same as the old piece?

Re:Darpa Seeks to Create Software Systems That Could Last 10... by dragonwriter on Wednesday April 29, 2015 @07:13PM (#9462443)
I would say it is the same plane in exactly the same sense that DARPA is interested in having the same computing system running for 100+ years; i.e., they want something that can be in use, maintained, adapted to current needs by incremental updates, without every being tossed out all at once and replaced with a from-scratch new system.
Re:Darpa Seeks to Create Software Systems That Could Last 10... by SomeStupidPoint on Wednesday April 29, 2015 @11:31PM (#9463204)
There's a second reason I think that they're phrasing it as "long term" computation, but their requirements discuss formal verification in some detail.

It's true that building up a large bank of formal proofs could take large amounts of super computer time -- eg, dealing with something the size of drivers or a kernel -- and that with the current upgrade cycle and manner that hardware is specified, this might not be practical.

However, if you developed a framework for your proofs that generated a proof generator which generated a final proof easily when given a schematic of the hardware to execute the algorithm and made such a framework general enough that you can use it to make schematics when you upgrade your hardware, then your time spent making the proof generator isn't wasted every hardware cycle.

Once your proof system isn't reset every time you upgrade hardware, it begins to make more sense to develop things like your firewall, IDS, routing, crypto, etc code in terms of formally verified systems (that generate machine specific code in automatic ways).

When you combine this project with other DARPA cybersecurity projects, it becomes clear that DARPA is looking to transition key infrastructure in to formally verified, machine upgradable systems in preparation of high-speed automated cyberwarfare. (Think the difference between HFT and a guy on the E-trade website applied to scanning software for vulnerabilities and generating things like Flame automatically.)

tl;dr: DARPA's plan for the next 10-20 years seems to be that they're afraid of AIs beating them at cyberwar, and this is a piece of their solution.