Cloud Native Computing Foundation Online Programs, 18 Aug 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: An Armful of Clusters: Run Kubernetes Workloads More Efficiently

Description

Arm processors are disrupting the computer industry on many fronts. First, they took the mobile industry and now they are storming PCs and data centers. With an excellent performance per watt ratio and affordable prices, they look like an excellent choice for sustainability and efficiency.

Migrating your Kubernetes clusters to Arm processors is a common way to take advantage of these benefits, but there are several considerations and obstacles that you need to be prepared to overcome:

Should you choose pure or hybrid Arm clusters with x86 clusters?
- Can Kubernetes run smoothly on Arm machines?
- Are your apps ready to run on Arm processors?
- Are your build pipelines ready to generate the multi-architecture images?
- Is your registry ready to manage several architectures for the same images?

In this talk, we help you run your workloads in a more efficient and sustainable way on the Arm architecture by using different solutions and strategies with open-source applications, such as Harbor.

You will also hear practical tips and tricks about:
- How to take advantage of the compatibility of different languages or technologies.
- How to adequately build pipelines to generate images for different architectures.
- How to properly store and use the images.
- What happens when one or more of my apps can’t run on Arm?

About the speaker

Carlos has more than 15 years of experience in IT, currently focused on the cloud native ecosystem. He tried once to explain Kubernetes to his wife and then both didn’t understand it.

Currently Technical Product Marketing Engineer at Giant Swarm. Father of three. Technology, video games, movies and music are his hobbies in his (scarce) free time.

More about multi-arch cloud: https://www.arm.com/campaigns/multi-arch-cloud-infrastructure

○○○ Further Arm Software Developer Resources ○○○

- Twitter: https://twitter.com/ArmSoftwareDev
- Reddit: https://www.reddit.com/r/ArmSoftwareDev/
- GitHub: https://github.com/arm-software
- Discord: https://discord.gg/armsoftwaredev
- Arm Developer Program: https://www.arm.com/resources/developer-program
- Arm Developer Hub: https://www.arm.com/developer-hub/servers-and-cloud-computing

A

Right, I think that should be enough time so welcome everyone. Thank you very much for joining, um got a great webinar today, all about kubernetes and how they run on arm very exciting, so I'm not hugely familiar with so I'm. Looking forward to learning um yeah just a few things before we kick off, um we have a chat function on the right hand, side of the screen, so throughout the webinar. If you've got any questions, you'd like to post to Carlos, please do he'll, um you know, stop sometimes and answer them or.

A

If not, we can wait to the end of the presentation, then we'll go through.

B

A

um Yeah, this meeting is also being recorded, so you know, if anything's you know want to go back and revisit. You can watch it afterwards and share it around so yeah without further Ado I'd like to say, welcome wild guest, uh Carlos, arrilla of giant swarm who's, a technical product, marketing engineer um and yeah so Carlos great to meet you and please yeah. Go.

B

Ahead welcome everyone. Thank you for giving me the chance to tag here and I am Carlos. I I work at giantram as a marketing as Josh very well said. uh Just a brief presentation about me. I have more than 15 years of experience in ID I've done a little bit of everything since I've been a iot devops and now I'm more specialized on kubernetes I've always been an open source.

B

Enthusiast, and here you have my first computer as Sinclair Tech Spectrum, very nice computer, a little bit about me, I, like all these things, Sports Video, Games, books, music and then Star Wars and everything like a geek I would say- and these are my three little helpers to the back kubernetes clusters and they're very very helpful.

B

A this is the agenda for today we'll talk about a little bit about arm, what's arm how they they was born in and what are the main advantages of the arm processors. Then I will talk about kubernetes a bit and how we can run kubernetes in arm and then uh all the applications that run on top of kubernetes how we can do all the conversions that will allow us to run them in in arm clusters and then questions and so on.

B

Hey I, don't mind if you have any question to stop any moment and answer if it's meaningful, if it's too different I can leave it for the end, but don't don't hesitate to ask questions anytime, you want so, let's start with some contests about arm a 1985 was a nice year. I was already here, I, don't know if most of you were here but Norway won. Eurovision come on the Spain I'm from Spain.

B

If I haven't said it a join eec which is current EU, and there were many very good films, uh Super Mario, Bros, Windows, 1.1, 1.0, nice and that year Acorn creates the first time processor. A the first version was something like this: a co-processor that worked in some special tax for uh another processor. But then they created a full processor and they they.

B

The architecture of the processor has evolved over time with many different versions like av3 in this computer and one of the most successful on computers, which is the Raspberry Pi, a it's already, some years old, but it's still rocking and very well known, and now we have Army in data centers too, like graviton. We have in in all the main club providers. They have the the arm machines.

B

So now it's very easy to use Arm processors, and the best example is that probably all the phones you have in your pockets will have arm processors, uh so it's they are everywhere.

B

The main advantages is that the architecture is made for Simplicity, so they are more efficient and especially interesting for mobile, because you know batteries are a very important and then the the design of the r processor was multi-purpose. So there are different cores a suite for different tasks, and that makes a very easy to customize the performance that you have and get a very good performance per dollar, and now they are comparable in the years ago.

B

They say that okay, mobiles are not so powerful, they are less powerful than desktop or laptop computers or x86 processors. But now performance is comparable, and sometimes we can say it's even a little bit better and one of the keys of the success of this iron platform was the licensing mode, because anyone can license the core and use them to make a customized SOC. So you can build your own phone computer or wherever around this course, so you can customize to make everything you want a um focus on performance focus on efficiency or mix.

B

So this allows all the computer and phone makers to to customize this these processors to make a future day one.

B

So the main two filters we are focusing today are power, efficiency and lower cost. That comes from Simplicity, and this is very, very important. uh We will see that the reason to use Arm processor is basically these two and let's talk about kubernetes, because uh kubernetes. This is a complex thing.

B

I always say that I once try to explain kubernetes to my wife and then we both didn't understand kubernetes, um but I will try, because you know we're here: uh kubernetes is a system that is designed to run a were lost to be resilient and to have high availability. It's it's designed for scalability. So the the architecture is something like this. You have some machines running a control plane that will take care of scheduling and everything that happens in the cluster, and then you have some notes that will run the applications themselves.

B

uh This is designed to run containers. So this is one of the systems that started the cloud native and.

B

It's it's been very, very important for the development or all these Technologies, um but many of you will have heard. Oh, but this super easy to install kubernetes cluster. You just take four or five Raspberry Pi's and install one of these softwares that works almost out of the box, k3s micro kubernetes, and then you have arm cluster Okay. So, okay, thank you for listening. My job here is done. If you have any questions you can, you can tell me add in okay: wait! No! No! No!

B

Okay! We can have our clusters very easy, but we are not talking about that. That is for home for hobbies, but we are talking about real professional applications. So, let's see how real kubernetes looks at least to me so I say this is kubernetes control, plane notes, but only this one um won't get you anywhere. You need things running on top of it.

B

So once you install this, you need security, you need observability, you need networking and you need a way to deploy things here and you need to create external resources, maybe communicate with Cloud providers, databases a all. These applications need to run in the cluster too, and this is the most difficult part of kubernetes, not just having the cluster having a cluster is easy. This is the difficult part, and on top of it, there are hundreds of Clone in the cloud native landscape is huge.

B

There are a lot of applications there and all these need to run in arm clusters too. So to say, okay arm, cluster kubernetes cluster are ready for production. We need to be sure that everything we need to make the application accessible from the outside to make it reliable, are working.

B

uh Let's see how we can design our clusters and there are different strategies and I will name three that I see that are the my the more important um the first one is: okay, I have my cluster my control plane. My control plane will be working on x66 notes, so uh it's like before, but I will change my x86 notes by arm nodes, and these are notes will run the applications that I want, and the control plane will only just take care of things.

B

uh What's cool about this and the control plane notes are still x66.

B

So you don't need to change how you deploy or how to configure your control plane, and many people has different systems to spawn clusters to to add them to your infrastructure, and this allows you to change very few things and still work and all the workers knows adequa. This is important because in kubernetes you usually don't need to say, in which note do you want the application to run. Kubernetes will make that itself and if there are the nodes, are equal. It's easier for for the system to schedule everything. What's not so good.

B

uh There are things in kubernetes called demon sets which are Bots or applications that run in all the notes, and that means it.

B

It means they're running arm nodes and in x86 notes, because control plane at the end of the day or not. So this could be a possible issue if your image is not very well, if not ready to run both architectures and you need arm images for everything that runs in the worker nodes, a you are not so efficient because half or at least some of the nodes are still x86.

B

So you are not getting everything when is recommended to use this strategy when your applications of them have ARM version, then it is a good strategy and you don't want to modify how I how you are creating your clusters. You have a complex system in place and you don't want to modify a lot of things, so this could be an easier step, just create an old pool with arm nodes, and that's it second strategy is more bold and I would say, for me is the best.

B

If you can do it ER it's go for arm all the notes in the cluster will be armed control, plane and and working out. uh What's very good about this, all the notes are equal. This is uh an advantage because it's easier to forget about multi-arc images or things like that.

B

um You get the maximum efficiency, because everything is armed. You don't have any exact disease note, but it has some problems. The first one is that you need arm images for everything you you run in the cluster. That means everything, observability, networking, security, everything and you probably will need to build and maintain some images that are not publicly available for some of the applications.

B

Another question is a kubernetes is a very well tested application, but it's not so well tested in arm clusters, so uh you will be trading a known floor and there might be issues in the applications that haven't been debugged, because there was no reason to to do it.

B

This might happen uh when it's good to use this application. This this architecture is you you have all your apps with an armed version, so it's very easy to go full arm.

B

uh You are willing to change how you deploy your clusters, and so, if you are flexible, you want to change, and you need to do some work to deploy all your clusters with arm and that's it and you want everything. I want all the cost. Saving I want all the efficiency. I want everything performance.

B

Then you need to go this this way and then a different strategy would be a half mixed notes uh in this strategy. I, don't really care if the control plane is arm, rx66 will depend on uh things we already said about this, but we will have mix uh notes. We have node pools with arm processors. We have node pools with x86 processors.

B

What's good about this, we can decide the control plane if we want to run it in arm or an excites, and we can run apps with no arm version. This is the most important and I would say this is the only reason strategy is like I need to use this specific application that won't run in our processors.

B

This is what we have. This strategy has many issues possible issues with them on sets, because you need to run them in all the nodes that would be problematic. Everything will need to be multi-arc, which is more complex and there's some one more problem when you use kubernetes.

B

This is a share resource, so you need to set request or unlimited of resources that you are using in the cluster and, as you don't know, the performance of the application in both architectures and it very well might be different and you might need to set different limits, and this is a problem because this can't be done with current kubernetes a resources.

B

So this could be a problem and we need to manage what, as I said before, Oh having all the knows equals an advantage, because it's very easy to schedule. You just send applications to two nodes indifferently, but now, if we have different type of Knob, we need to take care of where we are running everything- and this is complex. There are several mechanisms in kubernetes like North Affinity that you can say. Oh you run this application in exit.

B

This is notes this one in arm nodes, but you are complicating things and when you complicate things there are more chance for failure, errors and problems, and this again you don't have all the arm processors. So it's more efficient, it's less efficient, you don't have it all, and this is the only reason to use it when I need to use an app that doesn't have any a or any any way to run in our processor.

B

A there's one important thing about kubernetes: that is a you always have to test everything, because uh all the applications I use in my day-to-day in kubernetes probably are different that you use.

B

So my experience will be relevant in some level, but uh the only way to be sure that everything you are doing in kubernetes will work on arm is trying yourself is making a test environment is a running things and see where they break or if they work in the development process in in Cloud native I separated three different steps where this migration might be relevant. First is the compilation we are.

B

If we are used to compile everything for x-ray disease, we will need to put in place some way to compile the application for arm processors, and then there is the deployment or the CI CD by lines that you use to create all the artifacts that you need to deploy. The applications in the cluster a test, a everything that happens between my code, is ready and compiled.

B

I have my binary and everything is running in the cluster and there is another relevant part of the system which is not part of the cluster per se, but is very well related, which is the image registry. It's where you are storing your images and, in my experience, I have um very I I, wouldn't say very much very much, but quite a few issues with multi-arc in the Registries. So this uh this is something that you need to take care.

B

You need to research and be sure that it speaks so companion, Forum nowadays, I think this is pretty much solved and there are a lot of multi-arg ways to compile in all the languages.

B

There are still some languages that are more complex than others, but I would say if you are using Java, for example, there is a I would say pretty immature arm jbm because basically Android- and this thing is based on feeling, which is very similar to Java, a all the binaries binary, binary, compile language like C, go and others are very well uh resulted in a lot of cross, compilers and things to to do. And if you use an interpreted language, they will probably a very well uh a very good and mature interpreter for your language.

B

So I would say this is probably the easiest part. You will need your research. You will need to be sure that you are using the correct tools, but this is a matter of trying and doing it I'm more worried about cicd pipelines, because this is a complex field. There are a lot of things that could happen there and you need to be sure that you can do all these things.

B

First, you need to build your images when you are building your images, and nowadays there are a lot of arm computers out there, because there is a famous brand with a fruit that decided to go arm, uh but five years ago it was very strange to have an armed computer to compile things.

B

So you you needed gross compilation, you needed a virtualization, and that was an ideal.

B

uh So you need to be sure that you can do that that you can cross, compile that you can build multi-at images. There are several tools that allow you that Docker build decks. I think is the most common uh and it's I would say easy to use. But again you need to do your own research. You need to try it. You need to check that your applications work very well with with it so there's something else you need to do and then a in a perfect devops world after you build.

B

You need to test what you are building before, deploying of course, because you will test before deploying right, uh but in order to test arm applications, you need arm environment. You need arm machines, for example, if you are using a a self-host, a CI system, I'm thinking about Jenkins, or something like that.

B

You cannot arm workers and do this in Native machines. But if you are using a hosted service, this might be difficult.

B

uh I I know that some hosted systems like Circle CI support are machines right now, but that's something you need to check and I need to ensure, um and you don't need to remove your previous test like you still, while you are changing you'll, probably be months or even years using both architectures are an x86, so you need to be able to perform this in both architectures, which means that it will probably more expensive and will need more time and then deploy a you need to be ready to deploy in arm the Manifest need to have the architecture in there.

B

So it's easier for the for the notes to know which architecture are you using, and you need to be sure, as we said before, if we need something to clarify to the cluster which application goes in, which node you need to do that times in Toleration, no daffinity, all those things you need to do that. So there are several things you need to do in order to get everything ready, and this requires time a lot of time, and so this is not a matter of a week.

B

It's probably something like months and this process will be different depending on your applications. How many applications do you have and how experiences are they have teams with arm processors or with cicd tools and so on? So this is something you need to be very careful, and these can be expensive so, first and before you decide in oh, we are going full arm. You need to be sure that your increase of spending on processes like cicd, won't overcome the efficiency and the savings you are doing in our clusters and the third Point very interesting.

B

Is the registry I've seen a lot of things? I've seen and images that are retrieved with the Run architecture, so this is important right now, all the or at least the common Registries, with the latest version, support multi-arc and it's more or less well. Well, a result, but I know not. Everyone runs the latest version. Not everyone run the common Registries. So you need to be sure that you use a registry that is ready to support Monty Arc.

B

You have in mind too that multi-arc images are bigger, it's not twice as big, but they are bigger. So if you are using this, you will need to have more a space to store images. Network traffic will be higher, so you need to have this in mind and you need to be very careful with the problem. I've seen Registries, sending their run image or not supporting multi-art, but container runtime is a thing too container.

B

Runtime is the program that runs in the node that is actually running the container and I've seen things that made me get awake at night like and you say arm, but it downloads x80cc image, because, whatever all the container runtimes, they claim that they work with multi-arc, but I've seen issues uh I think in the oci specification, 1.0 multi-argis resolve, but anyway, I've seen issues in, for example, um acryo, which is a container engine uh as reason as six months. So you will. We need to check and don't trash blindly.

B

So the takeaway for this stock should be the cost. Saving can be very important and the power efficiency. uh When now that we are I, think the climate change is accelerating a a green ID. It's important and power efficiency, it's more important every day. So this is something we need to take into account, not only the money but the environment, but we need to have in mind that kubernetes in arm is not easy. I would say: kubernetes is not easy, so um moving to arm one make it easier.

B

If you want production grade clusters, you need to work a lot and to check a lot of applications and to try a lot of things. So this requires a lot of work and once the Clusters are working, adapting all the applications that we are doing is not trivial can be easier. It can be more difficult, but it's not trivial, so it requires some some you and you need to be very careful that maintaining all these duplicated pipelines and lines of development, a1b.

B

Worse than the power and the money that we are saving you see now, um so this is all for me. Thank you very much everyone, and if there is any question, uh I can answer anything well anything anything I know. Thank you very much.

A

Perfect, thank you very much Carlos. That was really really good um yeah, as cause mentioned. Please do put any questions in the chat to the right side. um I have a question myself. Actually Carlos watching through that I was thinking, say, I'm, someone who's, you know familiar with kubernetes and is.

B

A

Know tempted to move over to arm um but maybe concerned about any challenges or problems.

B

A

You say are some, you know what is the most difficult part of migrating to arm and do you have any tips for people and how they can? You know you know get past that.

B

I think, at the end of the day, you have control over your application, so migrating them to arm not very difficult, or at least it's a something under your control, but then in kubernetes you need to use a lot of applications that are third party that are open source and not all of them are ready and there is no quick way to have them ready. So, for example, I need to use Kong for API Gateway I need to use istio for networking psyllium a Prometheus Falco.

B

All these applications are open. Source are maintained by by several people and your priorities and their priorities might not be the same. So if they don't work, you might have a problem and that that is the more difficult thing to overcome like having all the I call. It auxiliary software or appliances that you use to make things work to make them ready for run. That's the most difficult part.

A

Ner, hopefully, that will help some help. Someone out and yeah I guess one more question: why is kubernetes like what is you know your elevated footprint? What's your favorite thing about it, you know in a quick, you know, there's.

B

A lot of people that things that oh kubernetes is the best, and there are a lot of.

A

B

Saying this is the wars or salt complexion I think they are both right.

B

The kubernetes system is designed for scalability and for reliability, and you need to use it when your applications require those. When you have a number of users that require special characteristics and special scalability, when you need like millions of users or specific requirements, if you have a Blog with 10 visitors per week, these are need kubernetes.

B

uh So you need to use it right. You need to use it when you really need it and that's where kubernetes writes, because it's it's allowed you to do things that are very, very useful for especially for scale and reliability.

B

That's that's the reason to use it. um If you try to use it for everything, you will probably fail.

A

Perfect well yeah.

B

A

Know good summary, um so yeah we've had no more questions coming in from uh from the from our watches, so I guess regardless, if there's um sort of, if people do want to get in contact with you afterwards of any burning questions, what would be the best way to get in touch.

B

um You can contact me on LinkedIn or uh here in the first slide this. This is my email, so feel free to drop me an email, and we can chat.

A

B

I'm easy to find yeah.

A

Excellent, well, that's great! Oh thanks once again Carlos that was really really good um and.

B

A

And thank you. Everyone for watching, I think we'll call it a day here. um Look forward to the next one and see you all soon, thanks again, all the better. Thank you. Bye.