Cloud Architecture (is the future)

Internet DNA Podcast

Dan and Abi continue to look at the technology that will shape our future and where you can invest time to stay afloat in that brave new world. From lego to data science taking in generative art and our usual dig at Netflix - film genre's really do need to be redesigned. How smart is smart?

Transcription

(this transcription is written by robots… so don’t be surprised!)

This week. This week we're going to discuss cloud architecture. Now, last time when we talked about cloud storage, I sang you

nailed Simon's clouds, but you know what? I'm not going to sing this time, but instead I thought you might like a bit of this silence, can you? No, no. I quite like the silence. Then it's a little fluffy clouds. Yeah. Anyway, I didn't go straight in with a to do to do little fluffy clouds, but I said I wasn't on a thing. I actually did the beginning with the cockroach, which took me back. I love that song. We used to play on high speed and rave to it before raving was a thing, but not for us at that point. That's how cool we are.

I'm 45 so I can dance to it. I know that I come across in the Dan and Abby show as not knowing very much at all and Dan, my guru, explains it all to me. Well in this episode, I really don't know anything at all. And other episodes I placed the question to get the answers and probably know quite a bit, but today I really needed educating. Last week we talked about the fact that if there's one thing as someone working in the future they should try and find out about looking to work with cloud architecture. So that was my point was that it was your appointment to me was, it is.

So if you imagined cloud services as bits of Lego I guess is the best way to look at it. So they have storage solutions, so you might have really high capacity or high speed or depending on what you're trying to do with the storage, they will have computers

that needs to be immediately accessible storage that can be archived, that

storage that is optimized for high throughput for things like if you were doing massive logging, you know where it just basically has to store it very, very fast. Other stuff might be very read intensive and there are usually lots of different services that you use for different reasons. So it's not like we have to, yeah,

I like chuck it. The only reason I say chunking is when I'm working at home. During the holidays, I have the worst internet in the world. We transfer can deal with it and I can only think that's because it does something very clever with little packets updating. I go, wow, you can do your staff too. That's my chunking realization.

So that's, we've talked about storage. You can have lots of different types of stories. You can have lots of different types of compute, so that can be anything from having a vps to really specialize computing that's full graphics or specialized computing that is really high memory or it's really high speeds IPC clock structure, right? Or it might not have a server at all. You might do serverless and all of these different services have different pros and cons and then you'll have other, like you might want mapping, you might want machine learning, you might want AI, you might want three d modeling in the old days if you would wanted to create something, an app of some sort, whether it's a web app or a native app, you would actually have to program all of this stuff together. Now you can literally just stitch it together from the different bits of Lego and build whatever you want really, really quickly, which means that you can prototype stuff, MVP, minimum viable product stuff really quickly and see if it works and because you only pay for what you need so you're not having to buy in a hundred servers, which most of the time you only use three and then occasionally you might need a hundred cause you're doing log processing because you're only paying for stuff as and when you need it.

And usually most cloud have quite a lot of free tier. You could probably run a website on AWS for free. In fact if you did it. Yeah,

which is Amazon web services. Or You could dob on GCP, which is Google clouds platform either. Yeah, this still sounds like storage. Why? Because it's all about getting the data and storing the data and retrieving the data. So it all sounds like dealing with stories. [inaudible] machine learning or anything. One of my Lego bits is going to be the bit that goes, I know how to be a customer service Chat Bot. Yes there is. And is that a cloud service?

Yeah. Cause there other things like Polly where you can say, right, what I want you to do is I want you to listen to a podcast and I want you to basically create a text version of it and then I can grab that text version and like I say, now I want it in Spanish or you've got to understand that obviously a lot of this works on data. I mean there's almost no app that won't have some data.

So in your architecture you've got storage, you've got some form of compute going on, which is running the programs. And then you've got all these apps, cloud apps. Is that what you'd call them? Micro services is what they're called. If I had a business that did everything like cause services, why? How do I get,

you might have some machine learning running over your sales catalog or your orders to see if it can the see any patterns. You might have some AI that's trying to predict what might happen or optimize the system. You might have mapping that's working out where people are so that your deliveries or your Justin time management of your scheduling is moving parts from one place to another place because it can predict, okay, there's going to be some order on that. You might have some three d modeling going on. Let's say you were building some sort of widget for bikes. You might want to do all the three d modeling and the stress testing and the flow die around through that. You might want to use something called [inaudible]

slow flow, which is what basically workflows.

Amazon uses a lot, so when you do an order rate, we'll start kicking off things, so we'll say right. That now needs to go through the correct warehouse. It needs to set up packing list. It needs to assign that picking this to either a machine or a person. Then it will go to the next power flow, which is once it's packed, then we need to schedule,

so this is why you get next day delivery because it's automated for the moment and you press buy on any day of the week, any time at night.

Yeah, and then they won't kick off the next part so it's not going to try and assign a van to someone until it's already packed and then it can assign the right van that's going to have the most optimized journey based on the [inaudible] will keep constantly updating that until the point where the vans actually pick up the boxes.

It is very clever as much as people don't like automating that game because it takes people's jobs. Although you still need the new people and the new jobs that look after the automation. It is very clever. It is. I mean that's why it's called smart. I know, and I'm not sounding particularly smart myself at the moment. More you hear about it and see what these companies are doing. Uber uses smart technology quite a lot, doesn't it? That's always Harold. It is.

They do Netflix do

what do you maybe use it for calculating how much people should pay and who is that?

Who's the nearest driver? I don't use Uber but you can do filters on that, can't you say? You can say right. Who's the nearest driver that meets my criteria? I only like women drivers.

We don't use Uber because there really is no Uber driver anywhere near us. Five to book a car. It would take a few days. Sadly

staring. It's constantly optimizing to try and get a, you are new Uber taxi as fast as possible and be to optimize the drivers so that they're already driving very short distances between pickups. You say, I want an Uber in 30 minutes. It's already worked out that actually if it puts that driver on that job, they will be next to you two minutes before you want picking up and that's why you but really only works in urban environments because you need a certain amount of through flow. The more journeys there are, the more optimized it is because it's got more to play with.

My brother is the human equivalent of that. He controls couriers and that's exactly what he is doing all the time.

And so basically what this could do is optimize some of that for him to say, well, you can just discount these four bikes because they're not going to, do you see what I mean? Just make his job easier.

Well, he's resigned, so maybe we just need to give them one of these bits of Lego to plug in. Go here. Here's CAS. Yeah, it's actually his cast 2.0 okay. That brings me into the next thing. When you say you just plug it in, you buy it from the cloud, you use what you need, you plug it in. Where do you plug it in?

Well, you build it in the cloud. So right at the beginning you create a thing called a VPC, which is a virtual private cloud, which basically means everything in that cloud is yours. And no one else's. No an LFC. No, not an LFC. [inaudible] not a little fluffy cat or Liverpool football club.

Well though they might use a v PC.

They probably do use a VPC.

Yeah. Refereeing is that using, yes,

Vir will be using elements of that. Be able to look at stuff. And certainly if you're in sports like Formula One, they're constantly using performance AI performance machine learning to understand how to make the race more optimize. When will they come in for a pit stop? How long did the tires last, how do we fulfill that? Enter back into the data so we can make better predictions going forward. But basically you create a virtual private cloud, which is your own little fluffy cloud and then you build things in there. So you say, right, I want a bit of storage, I need a compute to run the code. I'm going to connect some AI and ml. I'm going to do some ga work so I know where things are. And Yeah, the mapping, I'm going to have a little bit of AI that runs over it to constantly optimize that [inaudible] of website in the cloud.

Yeah. You're building all these things and when you say plugged in that lumps of code yet, so they have either API connections. So you say, I will ask the map for the following things. The API connection. An API is an application programming interface and what it allows you to do is to ask other programs for some structured data, usually in Jason format, so you can say to a map, give me the Geo coordinates of this and it will pass you back the Geo coordinates. Or it will say, give me a route based on points from point a to point B with a time based on traffic. For example, digital products talk to each other. Yeah, exactly. In a structured way so that digital products can't just come in and interfere with you. You only allow them to ask for specific data in specific ways. Cross products that just sort of waves in and go, absolutely.

I think I'm actually, I'm taking over your processes. Time in a bad mood. Yeah. So it's a way of allowing you to communicate with other people without, without the security problems that you would have with old style direct integrations. So it allows you to communicate between, and the other thing that's lovely about the cloud is once you've built all those things, you can say, right, these are all the parameters that I want of all the different bits. I'm going to save that into a file. And then you can say, right now I'm going to start another one in Japan and you just basically say create that in a VPC in Japan and it will just create all of the exactly the same thing again somewhere else scalable, which is the key part of the cloud, which is an idea called elasticity, which is instead of having a server, you basically say, I don't want any of my servers to run with more than 60% and when they do just spin up another identical server and share the load. And because all the choosing code, it can spin up a literally another server, another server, another server, another server, another gateway, some more AI. It can add more storage. Remove storage. So that's what's called elasticity. And it from the internet boom posting was prohibitively expensive because you just didn't know how much you needed. You couldn't have it falling over. You could be paying for wasted space. That elasticity

makes it much more cost effective and robust

and robust. And also imagine if your Netflix and your starting up, what you really don't want to have to do is buy 1 million servers today, not knowing. So you can just buy one and as your service grows it automatically scales out. It prevents a huge amount of upfront costs, which would've prevented companies from being able to start because it just wouldn't have the money to put them computers in place to build the systems because there would've been so much redundancy. So it allows people to be much more agile and start up costs much lower.

I want to take you back to something that you said at one of the Lego box you just threw in there was you add some AI, you add some machine learning. So I'm a business that does everything but isn't very digital yet. What do you mean? I just add some AI and add some machine learning.

Okay. So all of the cloud providers have a machine learning and AI. Leslie, you need to be a little bit aware of what AI does and machine learning does. You're not going to just say, I want machine learning. Well what does it meant to be learning? Oh, I don't know. We'll see that role. So obviously that you would need a reason for it to be there and then you just say, right, I want you to use these types of algorithms and I want you to work out these kinds of relationships. Or I want you to look at my orders and I want you to tell me things that I might not see any other way. So show me pattern matching or show me potentials or work out what are the propensities of somebody who has done that thing and that thing to end up somewhere else.

So machine learning is very much like a really, really good accountant. Sifting through some spreadsheets,

like a very good statistical analysis,

lacking data, data scientists showing you, yeah, information pattern that is emerging. And this is what I love about data at this visual side. So things that start emerging from this very dry numbers and then the AI, what's that doing?

So it's then starting to, let's say right now I can see these patterns. How would I most optimize the system to deliver somebody to whatever your end point is. So let's say your end point is I want someone to buy something. How do I optimize the journey for that person to get to the buy stage? So the machine learning has told you these are the likely funnels that they will go down. So now AI is starting to manipulate those funnels. So if you think AI is more predictive, it's trying to look at propensity is and move you around. Whereas machine learning is looking at existing data and saying what are the things that we can learn from [inaudible]?

But what's confusing as well is AI is learning, isn't it? So as it's doing these things, it's learning how to do it better and it's teaching itself.

That's machine learning really. If you think one of them is the data scientist and the other one is a sort of strategist, if you look at it that way. So one of them are saying this is the kind of things you can see are correlated. And the other one's saying, knowing those correlations, how do we alter the environment to take people to the places we want them to go to? Now that obviously there's a lot of mixture between AI and ml and their different sides of the same coin. If you think about it in that way,

but going back to last week when you were appalled that the guy in star trek who stood on the bridge and said [inaudible] response. Yes. Then the captain said, can you do anything about it? As in him saying, yes, yeah, AI would bypass that and it would know that it can do it. It could. It would do it. Now, would it learn that if it did it like this one time and something happened that would it do it slightly differently the next time

it would attempt to try all the different [inaudible]. So the machine learning would see what happened when I did that thing. The AI would try to predict which of the options would be best. So it's trying to say, given the situation I'm in and given what I know, which may not be complete, what do I think would be the best possible outcome. So you often get this in science fiction films where they go, there's only a 12% chance of success. That's what AI is trying to work out. What is the likelihood of success?

Hey, I now will go along and,

and do it. And then the machine learning will go, okay, we did that and this was the outcome. It wasn't what we expected. And that was because of this. So these are correlations that are, I need to bear in mind when I start doing my AI the next time around.

So they compliment each other. Yeah, exactly. That. And we've talked about narrow AI, like a car. It can only do one thing. And then broad AI smart. First of all, the smart stem for anything or is it just because it's meant to be smart? Like smart technology just means clever and so is smart AI or the combination of ml and AI is that, is that what smart is?

I think it's one of those words like fresh, which doesn't really mean anything

smart these days.

Like everything was turbo when we were young. It's got no legal meaning. Right. But people say I've got a smart TV. It's not really that smart. It's connected to the Internet so it can hiccup Netflix. I don't know how clever that really is.

Anything that's going to be the Internet of things that people call them smart.

Exactly. But it's come to mean connected to the Internet. So it is Iot. Yeah, exactly. Or enabled services. Not as smart as you might want it to be. Not like my TV goes high. I know what you like watching. [inaudible]

very nice. Well that's actually, I would say that's what they do.

Yeah, but they don't do that. Which is, we've talked about this before. I think one of the limitations of a lot of this is proper categorization and this is why you need data scientists to understand John was in categories are very limited way of understanding what people like. And that's why I think Netflix and Amazon prime or any of these things get it so wrong for some people because they look at the categories and genres and go all the people that these categories and genres like this sort of thing. But what they don't realize is actually he only likes a really tiny subset of that and how would we understand what that subset is and how that changes the way he specifically watches things,

those that it can be about the stories. It doesn't matter what the padding is, whether the story is set amongst politicians or in space or even in a costume drama. Yeah.

Queloz but he doesn't like explosions. Explosions. Yeah, she likes it.

Anyone dying. Yeah, there's a difficult one. Yeah,

exactly. So you'd have to start to really use some machine learning to really understand how to almost create categories and segments on the fly to start to say, okay, I can understand he likes thrillers but he only likes these type of thrillers and actually he only likes these types of thrillers because what he really likes about it is this and therefore we need to find a way of understanding this in our programming. John Rows and categories, this is sort of come out of where are all the jobs. So I think data science is one of those jobs where there probably will be quite a lot of,

I'm glad you brought it back to that. So you've said that what you really need to focus on is cloud architect, cloud systems, but let's say I am me with my business doing everything and you sake focus on cloud systems, which bit, what am I meant to do? I'm not meant to be the developer, surely which bit of cloud assistance should I be looking into?

That would really do depend on what you're looking to do. So you're looking at UI and UX. Then it would be how do I use the different components to improve the experience of the user? What bits of data don't know I need to place at what points of it? Yeah,

the whatever I'm interested in now try and extend that to see how that might be carried forward using, well I think I might start making smart art so that they don't even need to do the art. I just need to give it a few ideas and it probably would create the art just like my art anyway, maybe even better. So that's going to be my way of looking into the cloud and the future and AI machine learning

to understand what kind of visualizations sell well and then you could use the AI to generate better ones

and then I could give up the day job going completely be an artist

or a data scientist is what you probably end up turning into. Well because you would get so interested in how am I pulling the data? How is that changing

all the roads and being a data scientist?

No, not at all. Maybe that's what I'm interested in. Yeah.

Yeah. And what I am interested in is the visual side of the day to which you really can bring it to life. It is extraordinary and beautiful

and there's so much data now. I mean, there's millions and millions of API as you could use to generate your art with. That's quite an image.

I mean, I've worked with generative about, which has really fantastic is in the processing language and that is in itself, you set its parameters and it is creating art from whatever is set as its data and its parameters. So that's pretty interesting. We're going to have to call it a day though. Enjoy The sun. Speak to hi.

—

Dan & Abi work, talk & dream in tech. If you would like to discuss any speaking opportunity contact us.