 Computers keep changing the world, but their power and safety is limited by their rigid design. The T2 tile project works for bigger and safer computing using living systems principles. Follow our progress here on T Tuesday Updates. This is the 34th T Tuesday Update. Let's get into it. Last time it was about making ITCs, getting them delivered and making them. And the E-Series flasher trying to get the software load ready to go on the tile so that it would not be a tremendous waste of time to burn a software load that we would later have to disassemble tiles, or another one. I suspect I have to disassemble tiles and burn another one later, but I haven't fallen down to it yet. But it has kept me from building more than the Lotus that we had last time. Plenty more to do, so building more tiles is going to have to wait a little while. Last week it was about getting the paper that had done for the A-Life conference this summer in Newcastle, Pontein, in the UK, finished and shipped off in the camera-ready copy. That has now been done. I'll talk about it a little bit today. Main purpose of today, although there's lots to talk about, is to set a new goal. A good friend of mine was saying that, you know, it didn't matter that the first goal that we set at episode 10 to go for episode 20 ended up slipping a little bit. It was still good to have the goal. I totally agree. I'm going to set a new goal today. Okay, let's go. So, the paper. The paper is about a new cell membrane building on the C211 that was talked about last year in the A-Life conference in Tokyo in 2018. Now it's the C214 membrane three generations later. The title of the paper is Building a Survivable Pro Cell for a Corrosive Digital Environment. Now people on the inside are going to know what the corrosive digital environment we're talking about. It's Dreg and Res. And the membrane that we had last time was designed and tested in a completely empty universe, which was fine. It wasn't that the membrane survived well. It was that the membrane survived it all. Now time to move on a little bit. So, I'll show you some of the pictures from the paper. A lot of what I was doing last week was more citations and related work, which is one of the things that's really difficult. Because on the one hand, this thing is so crazy different than everything. It's hard to find stuff that says, if it's related to that, that makes you think it gives you the wrong impression about what else is true of this work. But there was found a bunch of stuff and, you know, related work is really important because it's sort of like the number one way that you're supposed to be identifying a community of people that are like-minded folks that you're in. I found a few people, especially Orhi and Black, unfortunately a lot of these references like Salsa are enables and all, they're sort of classic references from a while ago and they don't serve to build community network links as well. I gave it a try. I did find some cool stuff. The chemical networking protocols by Meyer and Chinden and others with the fraglets language, that is very cool. It doesn't immediately transfer to the movable feast because they use sort of a reaction vessel, which is kind of a 0D, where just these little fraglets kind of bounce off each other and do stuff. But it's very cool and it could generalize. So I did that and that was good for, you know, and so here it is. Here's the C211 membrane, last year's membrane on the left with a mob-like content inside and the new C214 membrane looks much the same in the membrane because it is much the same. It's got a different kind of guts, which is a lot more structured. It's like a routing grid going inside the membrane. The goal is for this to be a movable routing membrane. So it's got these HCs, which stands for hard site, which is short for hard cytoplasm, the stuff inside. It's also got soft site, although I don't think any of that is visible in that image. And the idea is that hard site can melt into soft site, migrate someplace else, and freeze up again, and that would be a way to allow a rigid sort of routing grid that would support the development of a local frame of reference, a local coordinate system so you could navigate around in it to actually move by melting and refreezing. That works a little bit, but it needs to work better, and that's not the primary focus of the paper. The primary focus of the paper is what do you do when you have dreg and res all around the outside of you sort of nibbling away at the edge of the membrane? And part of the answer is making the membrane better, changing the rules to correct cases that are not actually obeying the official rules of the membrane but are various forms of damage that we can fix by making additional little rules. You've got to be really careful that you don't cause cancer in the membrane by making changes which do not assume the membrane invariant and preserve the membrane invariant. So it did some changes to just improve the strength of the membrane. It did another set of changes where the outer membrane, when it sees a dreg, it pushes it away like that. I mean, according to the rules of dreg and res, you're not allowed to kill a dreg, dreg or sacred, even though they come around and would randomly destroy everything. But according to the way I took the ground rules in this paper, if you can't destroy a dreg, it doesn't mean you can't push it away. And so that's what one of the versions of the new membrane does. If it sees a dreg, it just moves it a little further away. If it comes back, it comes back. And you know, it's the movable feast. It's best effort. The dreg could get three events in a row and go, just burn its way all the way into the center of the cell and there's nothing a membrane can necessarily do about it. It's all best effort. The second step, the one that actually took it over the top as far as making the membrane much more survivable, was deploying a cloud of what I called cilia, like hairs, except they're not hairs. They're just individual cells floating around outside the membrane. And if they see a dreg in one direction and they see an outer membrane behind them, they use the outer membrane to say the direction that the cell is and they push the dreg even further away from the cell. And that improved the lifetime of the membrane tremendously. And here are the results. So this first curve here, I don't know if we can all see this. There's basically four curves. The one that reaches the halfway point, well, they're marked here. The 211, the 212, the 213 and the 214. The 211 and 212, I'm sorry, the 212 and 213 are intermediate ones. The 212 just has the membrane strengthening. The 213 has the membrane pushing away. And the 214 adds the cilia, pushing away even further. And all of these things are how long does it take until a membrane is breached? Just by blundering dreg coming around doesn't mean any harm. It's just what dreg does. But with the cilia, the average membrane, the median membrane out of 25 runs lasted longer than 500,000 apps and none of the other versions came anywhere close to that. So that's the primary result in the paper. There's a tremendous amount more to do with this thing. In particular, this cell does not move very well. And partly it's because of that interior, the hard sight grid is more rigid than the content was, the mob in the last year's model. But it's also because the world is full of dreg and res. And once it starts moving, you get a big collection of dreg and res and cilia at the leading edge of the cell tending to block it. So we really need to get the cilia to flow dreg and res around the cell membrane or something to actually make this thing move better. And that's future work. Don't know exactly when we're going to get to it, but that's the next step on this membrane. So that's the paper. I acknowledged y'all for bugging me about why am I not writing during the paper writing. For those folks who did, you know who you are. Thank you. And so that's it. Okay. And plenty more stuff to talk about. Let's just keep going. Need to save time at the end to talk about our new goal. So the key master, a whole lot of this stuff has been about organizing software so that changes can be made that can get pushed out to the grid. So right now we've got this crazy guy who's got a special two tone case that I made just for him. He's called the key master. The idea is this is meant to be at least one copy of the golden image. We hack on this thing, rebuild on this thing, and then figure out ways to push from the key master out to the grid. So this guy, for example, has a private key that the rest of the grid, well, the guys who have been updated, now recognize and will accept software from. So that's the key master. In addition, the intertile connectors that we ended up using, these expensive little connectors to build these things out of, because it was the only thing I could really be sure of. I have to say I actually like them. They're pretty smooth going in, smooth going out, or I may just be trying to rationalize the amount of price, you know, 60 cents versus 14 cents a pop of them. They were out of stock, so we couldn't finish the first order. They came back in stock earlier than expected, now 2,500 and something at Digi-Key. So that meant we could move on to getting the rest of these things assembled. These are the PDs, power and data. Also move on to get the DOs, the data only assembled. So this is what the DO circuit board looks like. It's got fewer connections because it doesn't connect the power pins. It just connects the data pins and the ground pins. It's got a cutout in the middle expressly to make it so that it doesn't, it won't take the same kind of connector as these guys. I work hard when I do these things. You know, this thing looks kind of symmetric, but if you look closely, this cutout box is not actually centered between the two. And the idea is if you have an inherent asymmetry, which we do in the connectors, they're key. They're supposed to go one way. What you try to do is carry that asymmetry through the rest of the design. And that's true in the intertile connectors too. You might not have ever noticed, but the northeast and northwest connectors are farther apart than the southwest and southeast connectors. Why? Because they don't need to be symmetric there, and by making them different, it makes it harder for you to accidentally plug the thing in upside down. So it did the same thing here with the cutout. Here's what the board looks like. I've now sent off an order for these things for the same folks, and they turned right around and bought 500 more of these connectors, of course, on my dime, which will be enough to finish the DPs and also to do the DOs, and we'll see when those come in. All right. When we tried to boot the first Lotus Live, there was a problem. Two of them didn't come up. I wasn't sure why. Further investigation, it appears it may actually not be my fault. It may actually be that a number of other people are running into this weird timeout problem. Occasionally with System D, this was from 2017. It's a little bit older, but there are other people acknowledging it as well, a race condition in System D. So it looks very much like what these guys ran into is what I'm running into. The official solution to it is to upgrade to a much newer version of Linux. I'm using a quite old version because the Beaglebone Green guys changed the rules that requires a substantial re-implementation that I don't want to take the time to do it. And I'm starting to wonder, you know, is there any chance I might be able to find somebody on Upwork or one of these contracting, you know, WorkGig, WorkSites, whatever it is, that I might be able to pay some money if they've got experience with Linux kernel modules and device trees and Beaglebone Linux stuff specifically. Man, I would pay them to figure out how to port the 4.4 Linux that I'm doing to a more current 4.9 and get the whole thing running again because I don't think I have the strength, but I'm not sure how to solve this race condition without it. We'll see. All right, what do you think this is? Here's another review. This was from last week's video when I was trying to plug the serial port connector into one of the tiles in the middle of the grid, and I actually managed to succeed in powering the guy next to him like that. Well, so now we've fixed that. So here it is. It's a little plastic gizmo that the guy gets, the serial port connector, that's this black thing with the wires popping out of it. Yeah, I ended up taping it into the 3D printed thing. I could glue it in, but this actually will come apart. I tried to design it so that it would snap together, but there really isn't enough clearance along this edge. It's the black, this edge, the connector's got to be pretty much open to manage to fit in. But it works great. So for example, now you just grab the little tab, it lines up, it's got a little flange, so it can't go in the wrong way, and it just, yes, very nice. And so now, you know, here it is, and it's easy. You grab it, you push it in, it's in. So that's, you know, this is settling. I posted an essay a while ago about exploring versus settling. I've been doing a lot of exploring in some sense, but the whole T2 tile project from one point of view is an act of settling. It's not settling meaning compromising, but settling meaning making the environment better for doing what you and your people want to do with it. Settlers. And so this kind of thing just makes life better, remove the pain points. There's a settler in me as well, even though I lean to the explorer. All right. And so now here is the Keymaster in operation. I made a change just, you know, particularly because I wanted it to, but partially just as a demo. So now when the stats display is up, instead of saying one, two, three, five, six, seven, it actually gives the names of the directions. It's not a big deal, but it was something that I could tell whether it was new or old. And so here's the Keymaster, who's connected up to one of the few other guys that I'm working with. It's still running the old version, one, two, three, five, six, seven. But after much toil and trouble, and there's still more to be done, now using the common data manager system, I managed to get a new version of the statistics display guy to move almost automatically. I still had to reboot the other tile myself to get the new one to take up, but it managed to pick up the new version. So that's just an existence proof on what is going to be this gigantic background distribution system. It's very slow, but the point is the movable feast is running the whole time in the foreground. This is not supposed to be fast. All right, so that's it for that. All right, so the goal is this. Let me go ahead and then come back to this. The goal is, if I can catch this thing, there it is, by the 40th episode. On July 9th, that's six weeks from today, we want the ring load we will have. The ring load us will be running, hanging on a wall, and it'll be ready for benchmarking to see how fast it can do drag physics. That is the goal. And I said, the previous goal, I said it with that same amount of determination, and I was wrong. I'm saying it with this amount of determination. It's from six weeks today, July 9th, ring load us run on a wall, ready for benchmark testing. And so that's it. That's the goal. There's a tremendous amount of stuff. There's every reason that I and we should be somewhat dubious about everything that has to happen between now and then, just to give you a flavor, because these things are going to be coming up over the next six weeks. We've got a lot more hardware work to do to get it onto a wall. The operating Linux system stuff, just beginning with the system in non-determinism. MFM needs a ton of work. We have to define data gathering software, and we have to figure out what the heck we're actually going to benchmark. That's not completely clear either. On the hardware side, we've got to build another 114 tiles. We've got to do the data only connectors. We've got to test the whole idea of putting multiple lotuses together with DOS in between them. We've got to do, we've got to build some kind of rigid frame to hold a lotus, because these things, oh, by the way, so I weighed these things. These things are about 219 grams a pop. That's about a half a pound. That means a lotus is almost 10 pounds, and the ring lotus is going to be 60 to 70 pounds. It's non-trivial. Now, on the one hand, it's nothing to be proud of, because the whole point is small and light, but as far as I'm concerned, this is the EMEAC. This is the first generation. It could weight 10 tons and take all the power of Niagara Falls to run. That's not the point. The point is what it does. So we should be all right with that. But it does mean we're going to need strength in these frames and in the interlotus connectors so that we can put this whole thing together and hang it from a crossbar or hang it from the ceiling, something like that, rather than imagining a picture hook on the wall is going to suffice. We've got a ton of work to do at the software level from figuring out how to get the CDM stuff more solid. It's mostly there. We also need some kind of broadcast channel to send messages like change simulation, power off, reboot, and so forth that want to be propagated through the entire grid. Pretty high priority flash traffic and then have a bit of a delay. So like, you know, say everybody power off in 10 seconds and have that flash through the entire grid then have the thing start, guys start shutting off eight or 10 seconds after that, and so on. Tons of work on the MFM level, that's sort of what I'm most afraid of or what really needs to be me and I've got to get some serious thinking time together to do it. More stuff. Actually how to run an experiment. What should we actually gather? What data to gather for a benchmark is not a small thing because you make decisions on these things and they define a pattern going forward for everybody else. We really would like to do them good. You know, not make them too easily gamed and that's a problem with benchmarks generally and they're always going to be a little vulnerable to that but we get a shot here, let's try to do it right. Alright, and that's it. Ring Lotus run in six weeks. It's going to happen. The next update will be out in a week. I really thank you guys. Ask me questions. Keep me honest. It's our little community. If people want to join, I'd love to have them but let's talk. I'll see you next week.