Will Binns
Will Binns avatar
Will Binns
@wbnns
Founder @mruspace • ran bitcoin.org, led teams @coinbase • editor of Mastering Bitcoin & Mastering Ethereum • software that outlives its makers
Will Binns avatar
Pinned
What does a machine do when there's nobody left to maintain it or tell it what to do? It stops. Every system we've ever built stops. Mru is a different answer: an operating system designed to run for a thousand years unattended.
Cast image embed
1
1
9
Will Binns avatar
Knowing exactly where you lose is how you know where you win Spare processors kept switched off beat us if they barely age while they wait. If they do age, we do up to 60% more work
Our simulator doesn't always say we win. If spare processors stay switched off and age slowly, keeping them cold does more work than our approach. We published that result next to the ones we like.
5
Will Binns avatar
We spend a lot of time on what a machine should do when its parts fail This is the harder question -- what it should do when its own update is wrong, and nobody is coming
Yesterday's FAA ground stop across all US airports may have started with an update. The cause isn't confirmed, but two of the three accounts of the outage so far point to one. Backups don't help when the same bad update reaches every copy at once. In Mru Field a new version comes in one copy at a time, and the old ver
quote image
1
1
4
Will Binns avatar
The best upgrade respects the work that's already there Mru Field works with the code you trust, not instead of it
Most remote equipment already runs code that works. Rewriting it to add fault tolerance is a project nobody has time for. Mru Field wraps the program you already have, in any language, and runs it as three checked copies. No rewrite.
1
3
Will Binns avatar
Yesterday's nationwide FAA ground stop was the right call... when planes can wait safely on the ground, stopping is the safe choice But hard to reach systems -- machines at sea, on the ice or in orbit -- don't get that option We make software that keeps them working through a failure instead of stopping for it
If you run equipment that's hard to reach, on the seafloor, on the ice, on a mountain or in orbit, we want to hear what breaks out there. The worst failures you've seen shape what we build next. contact@mru.space
1
8
Will Binns avatar
Most upgrades start with new hardware This one fits on what's already in the box, from the seafloor to orbit, and leaves 99% of the processor for the real work
Mru Flight's largest process uses about 2 MB of memory. The whole program is under 700 kB. On ESA's satellite in October, it used less than 1% of the processor, on hardware that was never designed with it in mind.
2
Will Binns avatar
A buoy on a fading battery faces the same choice: what to keep running, and in what order Getting that right is how a machine stays useful to the very end, on Earth or past Pluto
Voyager's power drops about 4 watts a year. Every year, engineers on Earth choose which instrument to switch off next, so the rest can keep going. A probe that goes farther will have to make those choices itself.
2
Will Binns avatar
We tried hard to make our own idea look bad Worst case we found: 11% more useful work. Best case: 68%. The simulator is open, so anyone can try their own numbers Still early. The goal is software that keeps machines running for millennia
A simulator is only as good as its guesses. So we tested ours against the numbers we're least sure of: how fast parts wear out, how often radiation flips a bit, how often failures come together. Across all of them, our approach did 1.11 to 1.68 times the useful work of standard triple voting, with no repairs.
quote image
1
6
Will Binns avatar
A sensor on the seafloor lives the same way between its once-a-year visits Mru helps it know what to do until the ship comes back
A spacecraft's hardware is fixed years before launch. Its software is the only part that can still learn, until the last day anyone can send an update. After that, it has to already know enough.
3
Will Binns avatar
Tests tell you it worked the times you tried. Proofs tell you it works the times you didn't think to try For a machine nobody can reach, we want both
The part of Mru that decides which copy to believe is small enough to check with math. Nine machine-checked proofs run on every change to it. Tests check examples. These proofs check every case in their scope.
1
1
Will Binns avatar
A volcano station matters most on the day it's hardest to reach We build software that helps it check its own work, so the people relying on it know which numbers to act on
Sensors on an active volcano are placed by people who hope not to go back soon. When one starts sending bad numbers, the choice is a dangerous climb or months of doubt about the readings that matter most.
1
Will Binns avatar
Some of the best engineering on Earth is a small team in California keeping a 49-year-old probe working We want every machine we send out to carry a little of that team with it
Voyager still has a team on Earth, 49 years after launch. When something breaks, they diagnose it and send a fix. Our simulator showed something hopeful: a team doesn't need to answer fast. A fix in one day or in six months kept the mission just as healthy. What counts is that someone is there. We build software that
1
4
Will Binns avatar
We build software for machines that will outlive everyone who built them So we write it like a letter to someone we'll never meet: plain, open, and easy to follow
A probe launched in our lifetime could still be working when our great-grandchildren are old. Nobody alive then will have built it. All they'll have is what it sends home, and the code we left them to understand it.
1
7
Will Binns avatar
Voyager 1, the farthest thing humans have ever built, is about to reach one light-day from Earth. Sirius is over 3,000 light-days away Fly there yourself and you'll see why we build for centuries
Mru 2049 is our open-source flight simulator: fly from Earth to Sirius through 109,389 real stars at their real distances. Fly far enough and the whole solar system shrinks to a single star behind you.
quote image
1
9
Will Binns avatar
Early days of a thousand-year job -- say exactly what your software will do, then do it in outer space 3/3 so far
Before each of our three runs in orbit, we wrote down exactly what should happen. Each time, it did. For a machine nobody can reach, being predictable is the whole point.
1