Samantha Coyle: Sorry, Neethu… [laughter] Yeah, I’m just gonna say microservices-based, Dockerized solutions are the future, I feel, because we are moving into an Internet of Things era, where more and more of these AI machine learning models are going to get integrated in the IoT field, and you need to have microservices-based, containerized solutions… And for microservices, I’ve been saying Golang is the language to develop on. Yeah.

Corey Sanders: I think that this is one of the challenges that actually we as an industry have with serverless. Serverless is fantastic when it comes to IoT, as you mentioned, and there’s going to be a core capability there, because it really breeds that scenario, right? Like, you’ve got your edge solution, you’ve got your IoT solution; when you need it, it fires, and when you don’t, it’s quiet.

David Kanter: Yeah, absolutely. So the mission of MLCommons is really making machine learning better for everyone. So I think of our goal as being how do we stimulate innovation in ML in a way that really benefits society and the whole world.

[04:22] So we were started– it was a very informal collaboration starting in 2018, but we actually formed a non-profit in 2020 that I am the founder and executive director of. We’ve got an amazing team, and it’s an industry consortium, so we’re bringing together a lot of heavyweights in the ML world from the systems side, from the software side, from the cloud side, who are all focused on this.

That mission of making ML better for the whole world - we have three real pillars to that. One is benchmarks and metrics. And so the first thing that a lot of people know us for is MLPerf, which are the industry standard benchmarks for the speed of training neural networks and doing inference. Actually, one of the things that’s been really cool since we’ve talked is– you know, I think when we first talked, we might have only had two of the benchmarks in our suite up and running. And we’re now, what I like to say is covering from microwatts to megawatts. So the smallest systems we’ve measured performance on are maybe tens of microwatts, really deeply embedded IoT devices, up to the world’s number one supercomputer, which is 20 megawatts, and everything in between. And so you can see what’s really cool.

I have some slides that I use in Keynote saying, “Hey, if you look at the two and a half years we’ve been around, you would expect that just through Moore’s law alone, ML solutions would be about 2,5x faster.” And that’s great, right? We want to make your life easier, Daniel, when you’re doing the work… But if you look at the MLPerf data that we have, actually, it’s more like 16x to 30x faster. And so it’s really cool to see when you get benchmarks and when you get metrics and everyone starts really rowing in the right direction, the kind of momentum you can get in the industry. So that’s one example of ways we can make ML better for everyone, is just being able to train bigger models and do it faster, and really bring those capabilities out.

The second pillar is the one that brings me here is our datasets. And I like to think of datasets as really being the raw ingredient for ML. If the industrial revolution was powered by iron and coal - not the most environmentally stuff - we’re really powered by data, right? And I think at SIL, you guys understand that just as well as everyone else; if you want to start talking in new languages or working with new texts, you can’t get from English to Urdu without some new data, right? And bringing large scale, open, nicely curated data is a huge boon for the industry, because even at some of the biggest shops in the world, places like Google or Amazon, their researchers want to use public data because the whole point is that you’re sharing techniques that everyone can use in driving the industry forward.

George Neville-Neil: [laughs] That’s right. You’re not worried about your manager walking down behind you. But yeah, the way people now pick up new skills is through these video courses. So that’s not me; I’m very much a reader, I wanna be left alone with a book, or a paper, or something like that, and a little peace and quiet, in a reading chair… But yeah, there are people who watch these things, and there’s a ton of content.

So the most important thing is get outside your box, find something that’s interesting still related to technology, that you think personally you could like… You know, when I first went into computer science back in the ’90s, it was one of the big uptakes of CS students in colleges, because they all realized they could make a lot of money.

So when I got to school, we had 300 undergraduates in my class studying computer science, and when I graduated, there were 89 of us. Because it turns out that most people don’t really enjoy what we do, and you have to be motivated. Back to an earlier question about artists - the same thing is true about programmers. I mean, yes, there are people who come in every day and crank out some amount of code on something that they don’t really care about… But I find those coders are not people I wanna work with. The people I wanna work with are the people who are excited about the technology they’re working on, they’re excited about the thing they’re building, or whatever it is… So the only way to broaden your interests is to broaden them in a way that works well with something you’re interested in.

[01:12:04.04] So I’m not gonna go from being a systems person to working on AI. I might work on the performance implications of the low-level code of that, but I’m not gonna sit around, trying to think up solutions to problems in AI. That’s too far of a jump. But certainly, within the systems area you can look at various operating systems, you can look at various chips, you can look at various applications.

IoT is a classic one of these now, where – we’re probably all doomed, but in the IoT world you’ve got computer technology applied to all kinds of different problems. You’ve got medical, you’ve got people doing power distribution… So if you know one component of the technology, you can be like “Well, I can pick up some other component of it by looking at the way someone else applies it.”

Paul Dix: I think it’s just a really nice fit to something general. The data model of Influx is obviously of a measurement name. You have a set of tags which are key value pairs, where the values are strings, and those are basically dimensions that you can slice and dice the data on. So a measurement name could be like, you know, grill temperature, and the tag information could be not just one barbecue, but many barbecues in many different locations, for many different people that are owners, or whatever.

And then the actual series data, which in Influx it’s what I call a field set, which is a field name, which would be just like a value, or something else, or it could be like inner temperature/outer temperature, or like a temperature for each grill… And then the value itself, and then finally a timestamp.

So it just so happens that that structure, that kind of schema works really well for IoT data, because it’s very common that people who are tracking sensor data want to slice and dice the data along those lines. But it’s also very common for server data as well, right? You want to see for this service how are things performing, or within this region how are things doing, or for this specific host.

I think the one other thing that is I think kind of unique with Influx and some of the other older time series solutions is that from day one we wanted it to be useful not just for what I call regular time series, which is measurements collected at fixed intervals, so like once a minute, but also irregular time series, which are like event-driven series. So that can be the response time for an individual request to an API, it could be a trade in the stock market, or stuff like that. Or a container spinning up or spinning down, that kind of stuff.

Mark Nadal: Three things. First off, if you don’t wanna do your own startup but you wanna advance your career and still do open source at the company that you’re at, I encourage you to just take a lot of initiative, find those problems, create solutions and then work with your company. You don’t think that you have to be a Silicon Valley startup company to have a lot of success on your own, or to be very happy with what you’re doing. So a lot of encouragement to people out there that you can, with your day job being a programmer, still make a lot of headway into this.

The second thing was what I already mentioned with Vue.js - if you’re big enough that you can go public and be popular, then you can take the Patreon route. The third thing here - the approach I’ve taken - is to build an open source community around your product. Everybody is going to use you for free, but then pay very specific attention to developers in that community that are building products on top of you. Right now, we have people doing everything from data visualization of IoT stuff, to virtual reality, web VR games, to your traditional Trello, social network apps, etc. And look for the people that are in your community that are developing a solution, a product that has big-name buyers that would bite, and that they have the connections to those big name buyers.

Partner with those people, potentially do a lot of free work - I’ve been doing a ton of free work for the last ten months - and make strategic bets, because that is going to be your partnership as a solution, as a product. You gotta convince them, and basically say “Hey, we’re built on your technology, so we have to use you, so why don’t we come together and as a consortium or as a partnership, or why don’t we, Gun, just hire you, and you bring the technology, the product, the solution that you’ve developed as one of our own, and you act as the ambassador out to these customers that you already have?”

[56:09] That’s the strategy I’m taking - I paid a lot of attention to the open source community and I’m trying to partner with the people that have the solution that these big companies need, and that solution can only be built on top of Gun. Then you capitalize that and crack the market open.

Daniel Whitenack: Yeah. And when I teach classes or workshops normally, the question comes up, like, when is an AI solution appropriate? Or under what conditions? And generally, I do think about that in terms of two things; one is scale… Even if a human, which I think a human would, in most cases, if they knew both languages, create a better translation… But if we have to translate 4 million sentences, the fact is that it’s going to take a human an incredible amount of time. And so there’s a scale factor, and then there’s also a complexity of the problem factor. So like, some problems, just by their very nature, even if they’re not at scale, are hard for a human to visualize. And maybe it’s combining a bunch of sensor data from IoT devices to determine when anomalies are happening in a network or in a manufacturing system or something. Those are like really complex problems; it’s hard for a human to actually parse all of that information and make a decision. Maybe they could retroactively, but in the moment, it’s difficult. So there’s the scale of complexity, but then also the scale of scale.

Divya: Yeah, so I listen to a podcast called Darknet Diaries. It’s really fascinating, because in my day-to-day I don’t deal with a lot of security issues as a frontend developer; that’s not something – I think in general, a lot of the times as a frontend developer you don’t really deal with a lot of security things… So I listen to it – it’s kind of a guilty pleasure, just to listen to what happens in the black/white-hat world. And there’s actually an episode in March, I think, where they talk about Samy… I didn’t realize, because I have not been updated on listening to podcasts, now that I don’t go anywhere.

It’s really cool, because you get to hear about different things, like security vulnerabilities that you might not have heard of… And just hear the inside story, because it’s not just talking about the vulnerability that happened, but it’s also the story of the person, the hacker themselves… And I think it humanizes them, because oftentimes when you read stories about things that are hacked… For instance, I read Krebs on Security, Krebs on Security – he’s a journalist, Brian Krebs, and he talks about the vulnerability and sort of the solution; what happened, as well as the resolution… So there’s not a lot of focus on the hacker themselves, and oftentimes – I mean, because he is a security researcher, the hacker is always shed in a poor light… And Darknet Diaries does sort of the opposite, where you empathize with the hacker a little, and you sort of feel bad for them, and you understand where they’re coming from, and you hear the back-story of how they started, what made them create a specific bug, and so on… Which I think are really cool.

[44:33] The other thing that – I mentioned Krebs on Security negatively, but I also read it quite positively… He posts really frequently. I think the one story that I was really hooked on was when he talked about Mirai, which was like a botnet on the IoT stuff… And it was really fascinating just to read about it. I don’t do a lot of IoT work, but he covered Mirai quite extensively, as well as talking about who the people were and how they created Mirai, and how Mirai affected a large portion of the U.S. And I think in general, if you read non-security things, Mirai was not talked about. No one knew that this was a thing, but in the security world it was pretty big. I find that really fascinating to read about.

And then the other thing also - Feross, you mentioned this earlier - is just vulnerabilities that happen in actual hardware… So because I read a lot of Brian Krebs, I’m paranoid about credit card machines and ATMs; so one of the things I do – because there’s a lot of these skimmers that people add onto the ATM machines, so when you put your ATM card into the card reader, it will skim your number, any of the information on the top… So people who come to take the thing can grab all of your information and basically clear your bank.

Oftentimes they come with a camera as well, so they can read your PIN… So they look at the card number and they see your PIN. So when I’m at an ATM now I always cover my number pad, and I always pull the card reader as well…

Pieter Cailliau: Yeah, so the difference is then you have graph native, you have an O(1) time complexity to go from one node to another node. So instead of an O(log N) in a relational database. And typically, you try to keep the data as close as possible, residing in the same key, effectively.

Then you’ve got RedisBloom, which is a set of probabilistic data structures, like a Bloom Filter, a Cuckoo Filter, we also have a Top-K… We’ve got RedisJSON, which is going to take your JSON document and it’s going to split it up inside a tree, so that in an atomic operation, in a large JSON document, you could for example append some data to an array, but you could increment a numeric value inside your JSON documents without having to fetch it and putting it back.

Then there is a RedisTimeSeries, which is the last one that went GA. It adds time series capabilities to Redis. It allows you to do, for example – in the role of IoT it’s very interesting; you’ve got lots of raw data, but you also would like to down-sample it. You could do that inside Redis already; there are many ways to do that. However, you have to write lots of client-side code. RedisTimeSeries comes with a toolbox to do aggregations over lots of samples, or to down-sample them, so you don’t keep all your raw data the further you go into history.

One more interesting module before we dig potentially into RedisAI is RedisGears. It’s interesting to mention, because RedisGears will be kind of like the serverless of Redis. You can do two things - you can do MapReduce type operations on top of your entire dataset, because Redis is a key-value store, so you can shard it, you can partition it… So if you want to run a query across all these shards, there was no clean solution for that. RedisGears will give that to you.

The second part is that it can also execute certain scripts based upon certain triggers or certain events. It could be for example you add something to a stream, or you change a certain key, but it could also be time-based; then you execute certain codes. This makes the entire multi-model story very beautiful. In our company, I’m the first one to say, if somebody says “Hey, multi-module/multi-model”, please be cautious about that. It’s not because we have multiple modules in a database that we’re a multi-model database… But RedisGears kind of allows you to do that. You have for example a stream of data, you’ve got some name change, for example… You could say “Hey, this script is going to react upon that new piece of data.” It’s going to say “Hey Redis, please update the name of this person”, but also “Hey, RedisGraph, make sure that graph person or that graph node is updated.” So it can effectively talk to all these modules, and then it becomes very interesting.

Chris Benson: Yeah, it’s always in the form – using it loosely as a service. I’ve seen web services used most often on server side, where you may not be constrained by your connectivity, and stuff… A lot of times though if your deployment target is an IoT device or a mobile device, you still have an API, but it’s really operating as a function, to use the phrase you’re using earlier that’s just – the API may not be a public API that your software component is using inside your group of software components that constitute your solution.

It doesn’t really matter, in my view, so long as that you are essentially following the best practices of the environment in which you’re coding and what your deployment target is made up of.

Suz Hinton: And I’ve always traditionally felt like the fun place, that weird edge case where we have this ragtag community… But IoT is actually a very serious thing that’s happening now in the Node space, and I’ve seen so many talks at this conference this week, and I was like “Oh, there’s other people out there, and they’re actually making money with this as a solution as well…” So it’s great that you brought that up earlier, because there’s a lot of communities like this that just don’t feel as though they’re basically as influential as they should be in order to have a spot at the table.

Ron Evans: Well, funny you should mention that… There’s really a few things that can make the big difference that currently are either completely lacking or are in the very early stage and need a lot more community involvement. The answer to everything is more community involvement, naturally. The first one is Bluetooth 4.0 a.k.a. Bluetooth Low Energy. This wireless standard is the core of device interactions with a whole category of different devices, and it’s really the fundament of the next generation of wireless IoT.

The best work that’s been done in this space has been by Roy Lee with the current lab’s BLE package. It is amazing how much work Roy has done pretty much almost entirely on his own. I mean, a few of us have helped out a little, little bit. He’s shouldering the technical debt for building out the entire BLE stack for Golang, essentially. Really remarkable work that he’s done, and needs a lot more support and help from the community.

Another one is running Golang on real-time operating systems like Zephyr or Mbed or FreeRTOS. There’s no reason why we shouldn’t be able to compile Golang and target it against an RTOS the same way that we’re targeting Windows or OS 10 or MIPS (Nine). It’s really a matter of somebody dedicating the time to… And there is a lot of documentation on how to build out this toolchain, it’s just that we need the community to rally behind that, because then that will allow Go to run on things like the Intel Curie, which is basically a dual processor - it’s got the microcontroller and it’s also got an ARC processor, so you can run real-time operating systems like Zephyr on it. Mbed, where you can run on a whole bunch of different ARM-based microcontrollers, so it’s a way of running Go on much less expensive, much more commoditized-type hardware. That would be a huge step forward.

And having some support for OpenCV, which is a computer vision package… You know, computer vision and machine learning - the other two pillars of the instrumented real world, right? We need the things that are the sensors giving us the data, and then we need these other software-based capabilities to analyze that data and figure out what it means, perhaps you wanna do something about it.

Our role the way that we see it as members of the Gobot team, we’re sort of the stewards of the de-facto low-level hardware interfaces for Go at this point. There have been a couple of movements, but whether or not this should be actually part of Go itself is very arguable, just because keeping Go very light and minimal and then adding these types of capabilities… So I don’t have a strong opinion about whether or not that should be.

[20:00] What I do have an opinion about is more if we as a community can sort of rally around creating a shared set of low-level hardware interfaces, to make it easier for implementers. We can then not just have Gobot, but potentially other projects that could utilize those same interfaces and share ideas. You know, a single solution or monoculture is not really sustainable. Because of our popularity, we view that our role is to be stewards of other interesting things that other people are doing, which may be in fact designed to replace what we’re doing at some point. Or they may be integrated into what we’re doing, since it’s all about the interfaces and helping the community create those.

It always just comes back down to “Let’s share ideas and see what happens.” But there’s a lot of exciting stuff going on as far as hardware-related activity. First of all, I made a list of some – just even this week. So I’m Los Angeles this week for the SCALE 15x conference. SCALE is the Southern California Linux Expo. it is North America’s largest community-run Linux expo, and it’s in its 15th year.