Kris Brandow: I think one of the interesting things about these network protocols is why did we develop TCP the way we did… I think a lot of that had to do with network congestion, and people just overloading networks. And when networks overload [unintelligible 00:49:41.28] that’s why we have window sizes, and all of these other things…

I think it’s interesting that we’ve gotten to a point now where those things have become more of a hindrance to us in some circumstances than they are a benefit. I think a really good example of that is the journey that HTTP has been on. We all started with HTTP/1, HTTP/1.1, where it was a text protocol and you couldn’t multiplex; and we had this head of line blocking problem, so you had to open multiple TCP connections to make to make browsers faster… And then we made HTTP/2, which added multiplexing, but it was still over a TCP connection, and we immediately ran into more head of line blocking problems, because TCP is ordered. So even if you have independent streams of things happening, if one gets blocked with processing or whatever, then all the strings get blocked, and this protocol that was supposed to be magically faster actually turned out to be really bad if you had a really crappy network…

And I think the other thing that we also started figuring out is that TCP works really great if you’re not moving, if you are a desktop, or you’re on Wi-Fi and you can establish those connections. But as soon as you’re hopping between Wi-Fi and a mobile network, now you have problems… Because with a mobile network they kind of make it appear as if you still have a TCP connection. But if you switch to Wi-Fi, your TCP connections are now gone and you have to reestablish them, which is super-expensive… So QUIC/HTTP/3 is now giving us the ability to move those connections between things, where you can go between a mobile network and Wi-Fi, and keep your connections alive. But it’s essentially just rebuilding everything we had in TCP on top of UDP, because TCP is too restrictive.

Adam Stacoviak: Just blatantly. Not even like network connection, but just more connection. Whenever you’re in conflict, the recipe to disperse and mitigate conflict is more connection, not less connection. So you would think that anybody in anything really would just thrive on more connection, so what you’re providing with sponsorship is just a greater social network, a greater work network, more connection to more people that know – whether you’re good or not doesn’t really matter. It’s your chance to now show up and prove or showcase who you are, and that connection can pay dividends over time.

Gerhard Lazu: [01:17:45.28] Yeah. So how fast is it? If we were to do here, you can see that right now, my download, I’m downloading 1.27 gigabits per second. And my network connection goes more. My network connection goes all the way to two gigabits. So right now I’m about 10,000, 11,000. So basically, Fastly is limiting my one connection - I mean, I say one connection - one IP, to about one point something gigabits per second. And maybe we could benchmark it elsewhere, but the point is you don’t want one user to use all your available bandwidth. So you need to apply some throttling.

So let me show you something interesting… If I, for example – let’s just do bench. So I do just bench, and if I do bench… Let’s do – remember bunny? We have bunnychangelog.com. Remember that? The other CDN? This is how the other CDN behaves, 1,700. So let’s go two. And I would like to go – let’s go 100,000 requests per second. Let’s see how that behaves.

Bert Hubert: Yeah, it is. And I come from a world where people also have to develop on networks with no internet connection. And it’s just basically impossible for many people to function there anymore, because you try to do something and - yeah, there’s no internet. There is no network. And the people that are in national security software, or encryption software, or that kind of stuff, intelligence agencies, they are actually in trouble, because the art of developing without a fully functioning network connection is dying out, because there are lots of people that need a runtime connected to the internet to get anything going. And even then, your software might be relying on third-party services. CDNs somewhere. And if you’re running this nuclear power plant where people say “Look, we’re not going to connect you to the internet”, it is becoming a vanishingly rare skill, of being able to build software that runs without a network connection.

Adam Stacoviak: [00:19:46.04] I think it goes back to my original answer with the other one - if you’re trying to be more out there and more of a well-known person, famous or infamous, then I think having a flashy GitHub profile, that shows off the things you care about, is going to be very helpful. Whether it’s for making friends, whether it’s making new network connections that lead to job opportunities, or a co-founder, or whatever it might be, then yeah. But if you want to be a developer who’s behind the scenes, not necessary. I mean, you may have a profile, but you may not care if it’s fleshed out in terms of, “Hey, this is Adam Stacoviak and what he cares about.”

Go look at my GitHub profile. It’s basically abandoned. I mean, I don’t do a lot of open source besides our own contributions, and that’s limited. I did do a lot of stuff in open source, but I’ve not done a lot in terms of code contributions.

Łukasz Langa: Right. So there were a few ways in which we worked around this obviously huge design limitation. So there are plenty of cases in which we’re confident we can drop the global interpreter lock for what is happening on a particular thread. So if your particular thread is doing some big NumPy computation that is only on the sides of numbers that are not shared as Python objects, you can drop the global interpreter lock in the time that this computation happens, so you can let other threads execute Python in that time. If your thread is waiting on a blocking network connection, it also does not need to hold the lock, because at that particular moment in time there is not going to be any Python code executing. So you can give that GIL to somebody else.

So there are special cases in which we were already sort of parallel more than not, but it is very implicit. It depends on us carefully finding all those special cases and enabling them. And in the end, it does not scale for a regular user application. So if you just write Python code that loops over things, and just executes some string concatenations, all these regular things that you expect from a Python program, most of the time you’re not going to be seeing any parallelism there.

Justin Garrison: [00:43:56.19] And, at least in my experience - I’ve never got a job that I didn’t have a network connection at a place.

Jerod Santo: Well, that’s what you find out about a lot of these things. They’re very easy to run. And then you’ve just got to make sure that they’re updated, or fix them. Getting it set up is the hard part. They run while they run. Things go wrong here or there… A network connection goes down… A time database gets out of sync… An update fails…

Jerod Santo: Right. So how much of that is file structure on disk, binary blob formats, that whole deal, layout of the data on disk, and how much of that is client server? Because it seems like when you say embedded we’re removing that client server connection, which oftentimes is a network connection, it could also be a socket… But that connection is going to be latency, right?

Ramin Mohammadi: Six months ago I was going to the AI Summit in Austin. I got to the airport and I was passing the TSA pre, where the TSA agent asked me to check my bag. It turned out that the machine picked up on this preworkout container which I had in my bag. The agent used this device on the box, and the result was positive. It was like “Oh man, I need to call for this special unit to come and check this.” I was like “Sure.”

Then this special unit with something like a Hazmat suit, they came and used a kit with a bunch of different reagent to more specifically test my pre workout. They started sampling from the pre-workout, and added to a bunch of these test tubes… Long story short, the agent was like “Yeah, you’re good.” He was like “This happened to my cousin also. These pre-workouts are causing false positives.” So we are like that special unit, with a bunch of different models trained to find and protect the dark PHI… While current tools in the market are like the first and the second machine, which leads to false positives or unknown false negatives.

At Tausight, we do see this problem as a personal problem. It’s our PHI that’s being targeted, and clearly the current tools in the market cannot protect it. So a key part of security also is that - you must do a complete and true assessment of all your risks and vulnerabilities to ePHI. It is so fundamental to what we need to do. And AI is such a critical piece of taking advantage of the newer technology. Solving that used to be a labor-intensive problem, and that could be much easier if you can define the scope of the problem, and you can have machine learning models which can run effectively and accurately in a calibrated manner. That’s what we do, we take advantage of the AI to find sensitive data.

I think we get to the point where risks and threats and vulnerabilities are going to be detected at the edge, using the AI… As opposed to “Gee, I have all these heuristic rules”, which is how we do lots of stuff today when it comes to recognizing patterns.

[00:24:04.10] So at Tausight we use AI, for example, to recognize when the sensitive data is in unstructured content. It doesn’t require us to say “Hey, there’s a keyword here, there’s another key word here.” That would be how you do heuristic programming. A combination of these three words must mean this; this combination of these four must mean this. Those – you never get to do all the rules; you will never get to the variability you need. So in our model, for example - one of our models, with about 50 million parameters, that can be set to recognize this stuff. You will never in years of programming get that much logic into your regex.

Now, the other main factor for us is the ability to run these models right at the edge, where the data is being created, emailed, printed, copied or faxed. We bring the AI to the data, rather than taking the data to the AI, which most of the current AI solutions do that. By doing so, we can ensure that our data is always protected, agnostic of hardware specs or network connections.

Louis Pilfold: The thing that’s really cool is a framework called Lustre, and Lustre started as being a – I guess it’s in the same sort of space as like Elm or React. It’s like a frontend framework, so it’s taken advantage of Gleam’s ability to compile to JavaScript, and then run in the browser. And then you can do your state updating stuff in the browser, etc. But then they extended it to – Hayleigh, the creator, she extended it to also be able to do server-side rendering.

[00:20:07.01] So you could use it in your backend, with Wisp, for example, to render your HTML templates, or that sort of thing. And then you could like mount bits of Lustre frontend code on top of those templates… And then with the latest version, she’s introduced server components, which is basically Elixir Live View. So using Lustre, you can have the best of both. You can say “Well, these bits we want to work really well, really snappy, not worry about how bad the network connection is. We can do that on the browser. While these bits, we want to have that really nice Lustre developer experience,be connected to the backend, and have all the data… Okay, we can use Server Components. And this stuff we want to be really simple, we’re going to server rendering.” You can have one ecosystem, write one code, and it just - every single possible way of rendering your frontend, it can do. It’s super-cool.

Johnny Boursiquot: It doesn’t have to be, right? I mean, I think a lot of times in my experience it’s team-dependent. Personally, on my teams – so concrete example. Rather than passing an instance of SQLDB, or something that implements it, like GORM, or SQLx, or whatever you want to use; rather than passing a value of that particular type, I’ll use a data access pattern instead. I’ll say “Okay, here’s a thing that is able to provide you the data you’re looking for”, whether that thing uses an in-memory data store, or whether it actually uses a network connection to make a database request; it doesn’t matter to you the thing that needs to say “Hey, DAO, can I get a list of users?” or whatever. That way, you’re not forced to then for every single thing that might need data, you’re not forced to then say “Well, I need to know how to handle a database connection when I initialize this particular struct” or whatever. You just pass on a DAO. That way that DAO, you can just have a test for that. If you want to have one where you’re dealing with database connections, that test is localized to just that thing, and then no other thing that uses it, no other service or struct or whatever it is needs to know how you get this particular kind of data. You just simply have a DAO that [unintelligible 00:26:28.11] data. So that’s a decision that I had to make, that “Hey, on our team, this is how we’re going to do this.” But I’ve seen it done differently in other teams. As long as you’re consistent, I’d rather the consistency than a hodgepodge different mixtures of styling approaches on the same project.

Gerhard Lazu: Okay. I think it’s that flexibility of getting a Talos node anywhere, and then connecting it, and connecting them all together… Which is really, really interesting when you look at them all together. How do you deal with unreliable network connections?

Richard Feldman: I would say there isn’t a focus on Web Assembly. So Roc compiles to either machine code or to Web Assembly. Actually, almost all the use cases today are not Web Assembly. The one that’s on the website is we have a web REPL. So you can try out the language right in the browser. And actually, the entire REPL is running in the browser, so you can like turn off your network connection and it still works. And that’s basically like – we have a little stripped-down version of the Roc compiler compile to WebAssembly, running in the browser. And then we also have, obviously, it compiles your Roc code to WebAssembly and then it runs it in the browser. So that’s like the main use case for it. But there’s plenty of stuff people can use WebAssembly for, so if people want to, they can do that.

Michael Quigley: Yeah, yeah. And the nice thing about the design too is you can take whatever layers make the most sense for what you’re trying to do – like, we generally ship the entire stack to customers, and they use the whole thing, but you could use just the mesh if you wanted to, you can… And then going up even higher, now that Ziti is starting to become mature, I’ve sort of pivoted my work and I work on an open source project called zrok now. And zrok was sort of a – the name, you can kind of tell what it was inspired by just the name. Most people are familiar with ngrok; that kind of had the… I was basically like “Go figure out what something like ngrok would look like on top of OpenZiti.” So I built zrok. And one of the cool things about zrok is that you look at tools like ngrok, and it’s awesome that you can share a private resource on a public URL, and let anyone use it, and you don’t have to open any security holes… All those things are great. But one of the cool things about Ziti is because we have this overlay network, and we control the addressing, and it’s all zero trust, we can do peer-to-peer across that network without requiring any of the peers to expose themselves publicly, or anything like that. All the peers can remain “dark”, but we can still have peer-to-peer communication. So that lets us do things like private sharing, where we can say “I want to share a network connection between two completely dark peers”, and the traffic will go across the overlay, across the internet, but none of that’s exposed to any of those things. All that data is end­-to-end encrypted, it’s all zero trust under the covers… So that’s one of the things that’s really cool about it. So zrok ends up becoming like a really good use case for showing people how you can use something like Ziti to build things that you couldn’t otherwise easily build.

Adrian Hesketh: Yeah, that is unfortunate. And actually - yeah, it is a mixed team, where all three of those editors are in play. But I think, actually, most of the teams that we work in - I think, Joe, you’d say the same - the editor, we kind of switch between editors a lot. So I think people really like the “Code With Me feature in the JetBrains stuff. And they see that that seems to work. If you have a dodgy network connection from time to time, it seems to be a little bit more resilient to that than VS Code… So I think we find ourselves using quite a few different editors. We practice a lot of mob programming at work, so there’s generally – sometimes we’ll all be on the same machine, or sometimes we’ll be taking turns, or different things. In practice, it turned out not to be a problem, but it’s definitely something we want to sort out, because I think every year there’s a go – what is it, like a questionnaire, with like a summary of what people are doing. I think it’s always like, VS Code is one of the biggest, and then it’s Go Land, and then it’s like Vim. But I think Vim has a surprising – what is it, like 20% usage? Quite high.

Andy Klein: …which - each server is its own server unit. It’s got its own intelligence, its own processor, its own 10G network connections, and whatever else. And so each one has its own boot drive as well. So that’s where we get them all from.

The boot drive for us does more than just boot the system. Like I mentioned earlier, it stores some of the smart stats on it for a period of time, it actually stores access logs and a few other things that we keep track of on a regular basis… Because one, there’s a compliance aspect to doing that, and then two, there’s just a maintenance aspect, and a debugging aspect. When something goes a little wonky, you want to be able to look through various logs. So we keep those for a period of time. And they get stored on that SSD as well, or the boot drive as well.

The SSD is - to be honest, we started using those because the price point came down to the point where we wanted to pay for it. [laughs]

Mike Riley: Now, if you buy these kits, which cost a little bit more, that include a power supply and a case, and maybe a few connectors, because not everybody has a USB-C to HDMI converter… So if you wanna spend a little bit more money for the kits that have all that typical stuff bundled in - that’s probably a better approach, especially for newcomers who aren’t aware of the fact that you’ve gotta have all these different attachments in order to fully utilize the Pi. If you don’t do that, then you will have to get yourself a USB to HDMI adapter. Or if you’ve got a newer monitor that actually accepts USB, you can get a USB to USB-C cable that will allow that signal to come through.

You’re gonna need power, so again, if you didn’t buy the Pi with a power adapter, you’re gonna have to make sure that you get the USB-C power adapter. In the earlier models of the Pi it was somewhat a proprietary choice on their part, on the part of Raspberry Pi Foundation… And it was not compatible with a few other USB-C power adapters. I know the newer ones have rectified that problem. But either way, you do probably wanna get yourself a power adapter and you wanna make sure that it’s gonna be Pi-compatible. So you look for your fully Pi 4 compatible USB-C power adapters. The Raspberry Pi Foundation actually sells one. It’s not that much more of an increase in price compared to some of the third-parties. So if you wanna stay safe, I recommend going the Raspberry PiFoundation’s version of the USB-C power adapter.

Then after that you’re gonna need a microSD card. You can get started with low-end, low-cost, a 32-gig, but if you’re like me and a lot of other Pi users, you’re gonna find out that that 32 gigs get consumed real quick, especially if you’ve got projects that are either looking at video, or you’re processing a lot of data, you got it for storage needs, what have you. So I usually go with a minimum of 128 gigs.

Now, you can get those SD cards for under $25 these days, even cheaper whenever they’re on sale. So the 128-gig is not that bad.

All to all though, by the end of the day, when you’ve got all those pieces of external accessories put together, that 8-gig Pi 4 is probably gonna cost you around $110… Which is still quite reasonable, considering the amount of compute power and the capability you’re getting out of this computer. But once you’ve got that all connected, then obviously you’re also gonna need a keyboard, probably a mouse if you’ve got the Raspberry Pi OS, which has its own GUI, which is Ubuntu-based… And then you’re going to have to have a monitor, so that you can see what you’re doing. Once you’ve got the Pi set up though, normally what I recommend is you try to get that on an SSH daemon as quickly as possible, so that way you don’t have to have that Pi connected to your keyboard, your mouse, your monitor. So you’ve got a network connection, whether that’s through your Ethernet wire connection, or you’ve already set it up for your Wi-Fi; it makes it much, much easier to just secure-shell into it and then be able to issue your commands that way.

[36:03] There’s a couple of really nice add-ons for something like an IDE, like Visual Studio Code, that allow you to really easily remotely connect to a device like a Pi over SSH, and be able to browse the files, be able to modify them, to update your code in almost real-time… And it’s almost like you’re working locally on the Pi, but in fact you’re doing it over this SSH connection. I talk about that briefly in the book as one of the recommended add-ons if you’re using Visual Studio Code for your Python code.

Ed Welch: I thought about this last night… [laughs] I’m not sure what it would look like to – I mean, you could do definitely something that catches it; where it’s tricky is how easy and how organic would it be to use… Because a lot of times you’re catching an error on a function, and you have no idea what the error was. The error might have been that the context was canceled while you were executing that function, and you’re just logging the error. You’re basically now thinking about wrapping all of your errors with this thing that would look for someone that’s better at this than I am.

As much as I love how well that context cancelation allows you to control over network connections and things, I find that it leaves a lot of times confusion around what piece of the puzzle was the one that actually said [unintelligible 00:21:57.15]

Adam Stacoviak: Just blatantly. Not even like network connection, but just more connection. Whenever you’re in conflict, the recipe to disperse and mitigate conflict is more connection, not less connection. So you would think that anybody in anything really would just thrive on more connection, so what you’re providing with sponsorship is just a greater social network, a greater work network, more connection to more people that know – whether you’re good or not doesn’t really matter. It’s your chance to now show up and prove or showcase who you are, and that connection can pay dividends over time.

Ryan Dahl: All the time. I mean, it depends on what you’re doing. Let’s say you’re writing a program like ESLint - this ESLint-like program doesn’t need to make outbound network connections. It doesn’t need to write to the disk. All it needs to do is read from the disk. So you would only enable the allow read in that case.

Kevin Ball: I’d love to dig in a little bit more on why it’s faster. You’ve mentioned a couple times that it is faster, and you have demos that show that… I was thinking about this and I was like “Okay, there’s a few different angles”, and I’d be curious to [unintelligible 00:33:12.16]

So one possibility that I was thinking about is you’ve essentially moved all the stuff that lived in the kernel into user space, and you’ve probably simplified the set of things it has to handle… So by virtue of not dipping down into kernel space, you don’t have any sort of process changing or anything, so that piece could be faster.

And then the other piece I was thinking about is you’re mostly simulating web stuff. So you could probably shortcut the network connections, so what would be a local host connection that’s going through your local network device, and HTTP, and all these other things you could probably put shortcuts in… And then there might be other things I haven’t even thought of, so I’m curious, what is it that’s making it so fast?

André Eriksson: I think that’s super-interesting. I really agree with your characterization of the different layers. I guess Encore is in many ways contrarian, in that I firmly believe that to really unlock much greater productivity, we actually need to look across these layers. Nowadays, most tools, as you pointed out, only operate in one layer. But when you actually bridge that gap, you end up getting something that is much more powerful. And what underpins Encore is really this belief that to really provide a better experience in all of these layers, we need to better understand what an application is doing. Because when we as developers are building an application, we have a mental picture in our heads of how it works, how everything fits together. And generally speaking, our tools do not share that understanding. They generally just think of it as code, and files, when you’re in an editor. And when you add this operations layer, suddenly we package everything into a container and treat it as a black box, where we have no idea how it all fits together. We can kind of try to figure it out based on network connections between different parts, and so on… But we’re really lacking this map of how everything fits together.

So what Encore is trying to do is really bridge that gap and build up a very detailed mental model of how your application fits together, and then try to improve on all of these layers. So that’s why we’re combining what is essentially a framework that’s just a way to get this sort of understanding, and then we use that understanding in the other layers to provide things like tracing, and so on… Because it turns out that the challenges with developer productivity - they don’t firmly fit into one layer or the other, and by looking across the hall, we can create a much better experience.