Cloud Native Computing Foundation KCD Amsterdam 2023, 6 Apr 2023

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Keynote - Crossing the chasm with multi arch by Cheryl Hung

Description

Cheryl Hung speaks to crossing the chasm with multi arch.

A

All right, let's see if this is gonna work, okay, so this is the last Talk of the day. So I really appreciate you staying around I'm going to keep this quite light and fun. Hopefully, nothing too serious. uh I have called this talk crossing the chasm with multi-art but, as I said, I want to keep it light and I also want to talk a little bit about some of my time inside cncf and some of the stories and some of the things that I I.

A

Remember from that, because I think it's a different Viewpoint from the community side, um and so my background is, uh if you look if you go to LinkedIn, this is what it will say so started out. It's Google as a c plus engineer, I moved into infrastructure, develop advocacy, I started a Meetup called Cloud native London, which has about 7 000 members and then 2018 I moved into cncf.

A

That's the kind of like very glossy story. um The truth of the matter is um I. Think I decided to join Google when I was a teenager, because I heard Google has free food and massages I genuinely did not really know what software engineering was at that point or why anyone would would do it, but I thought hey. This sounds like a cool company sounds like they're doing lots of fun stuff, um so I really enjoyed Google because I went. You know, studied computer science. Also I could get the free food and massages at Google.

A

um I spent five years working on Google Maps, various back-end features um and I left 2015, just as containers was coming up. So when I was looking around for what to do, I was like oh I've done. Borg inside Google I know how this works. This is the same thing, so I kind of just moved into infrastructure, because I happened to know what it was from five years at Google, I moved into developer advocacy at the same time again fairly early, like 2016, was quite early for developer.

A

Advocacy I moved into that because I was like Hey I want to be the sort of person who travels and like gets up on stage and does Keynotes that's kind of what I did I started: Cloud native London, the Meetup, because I was looking for the events to speak at and there was a cloud native Palace and a cloud native Berlin and there wasn't a cloud native London. So I was like ah whatever I'll just like start.

A

It put it on Meetup see if anyone shows up I picked a date that was about two months away: I had nothing, no speakers, no venue, no sponsors um and then more than 100 people showed up on the first one they're like okay cool, like there must be some some interest here and that's how I kind of moved into open source and community and then cncf 2018 um I had uh I had a job offer from AWS. At the time.

A

uh Does anyone who worked for AWS by the way, okay, I didn't want to work for AWS, so um I slacked, uh someone I, knew at cncf Chris, angik and I asked him like Hey I've got this offer from AWS, but is there anything interesting happening at cncf that you know you think I could do and he was like yeah we're two weeks away from hiring a director of ecosystem? You would be great you should you should join him.

A

So the whole point of me telling you that is that, like I kind of fell a lot into the open source infrastructure container World a little bit just by coincidence and good timing, a lot of it um and I want to sort of tell you sort of three particular times. I really remember from cncf.

A

First one is called after tntf. You could be. This was when I was interviewing with cncf I was about to I was waiting for a phone call from Dan Cohn. He was the executive director.

A

um So getting a bit nervous, you know butterflies in the stomach and phone rang. I picked it up. I said hi I'm Cheryl and he said hi Cheryl, I'm, Dan Cohn and after cncf you could be executive director of your own Foundation or you could be a VC or you could go work at one of our member companies and I was like I hadn't said anything I was like. What do you mean after cncf? I could be like.

A

This is a job interview, I'm supposed to be telling you why I'm supposed to be here right right, like did I did I mess up, did I already get the job and I didn't notice, uh but I really realized over time. That's how Dan thought about people. He was always like hey. How can I set you up?

A

So the next thing that you do you can have even bigger opportunities and do even greater things that you want to do and I've kind of tried to keep that in mind as I worked with Community with open source and try not to sort of say you know, can you do things today, but what things could people do in the future, that's even bigger and what they want to do?

A

The second second thing that I remember is when I was a little few months into cncf and I was getting my bearings trying to figure out what was the most important thing. That I should be spending my time on. So I wrote this long list of things that I could do and I took them to Dan and I said hey, which of these do you think, is the most important for me and he listened to me talk, and he said to me very very calmly.

A

You know Cheryl failure is an option and what he meant by that was. It doesn't really matter in a sense like you can't predict in advance which of these are going to work out or not, and failure is an option. So don't get too in your head about what is the perfect thing to do and trying to set everything up perfectly.

A

Just try a few different things and see which ones work and which ones don't and I was so grateful to him, especially in a new job, where you're so anxious to try and do the right thing that he said to me. Failure is an option, just don't stress too much and a lot of the things that actually failed. That I did at cncf. No one remembers them now, thankfully, but quite a few things that I did just didn't work out, and that was okay of the things that I did. That did work out.

A

Well, um people were in timing, mattered, the most and I think Dawn sort of some of the stuff dawn said also reflects upon this. So my favorite story of this is uh it's not something that's particularly well known, but it's called the cloud native maturity model um I had a consultant friend who said hi Joe, like we've got this maturity model that we use our consultancy. We help evaluate how far along the cloud native Journey these different organizations are. uh We think cncf Community might be interested in it. Why don't you publish it?

A

So I was like okay cool, but one company's opinion on this is not it's not really a community thing, so I filed it away and said: okay, we'll come back to it. A year later, a different friend or a different consultancy said: we've been working on a maturity model to help evaluate why organizations like how far organizations are along to adopting Cloud native, so I thought. Okay, like this sounds familiar, let me do a little bit of searching turns out.

A

A third friend of mine had also published a different maturity model, so I was like okay. This must be something interesting, let's form a group around it. Let's go and publish it, and this group ended up being called the cartographers group and they're looking at different ways of mapping out the journey to Cloud native I wanted to point this out, because there was probably there was a year, as I said between the first time. Someone said hey. This is something interesting.

A

We've got and oh there's enough other people around it that we can do something together and make it happen. So I think the value of like being at these kind of events is really about talking to people and finding out what stuff is happening now and then maybe it'll pan out in a few years. Maybe not, but that's how the things that I've done that have worked out have always come about so like what am I looking at now, so I work at arm.

A

Now the chip design company, let's say um and I want to sort of like arm, is almost founded in 1985. So here's some other things that happened in 1985. They found the Titanic the wreck of the Titanic George Michael's Careless Whisper was the number one song in that year. uh Nintendo published Super, Mario, Brothers, Windows, 1.0 came out and Back to the Future was the top grossing movie in that year.

A

Now in 2023 we live in a slightly different world, so connectivity and data wise we've got tons more data that we need to store and manage. We have 5G coming well, no it's already here, but 5G continues to increase, IIT continues to increase and then in terms of the trends, everyone knows. The markets are a little bit tricky time right now. Companies are looking for cost savings.

A

uh Moore's Law is coming to an end, so we can no longer double our performance easily as easily in the past, and then lots of companies also have corporate mandates and we should do the right thing to save money and be more sustained. So not save money save you know, CO2 be more efficient with our use of energy and save the planet.

A

So in the context of that, arm has been really interesting because um has focused on a couple of things. One is that it's a risk architecture so arm stands for advanced risk machines, and that means that it's focused on power efficiency, which is particularly good for mobile devices and right now for data centers.

A

Secondly, it's multi-purpose, so you can create different cores specialized for different tasks, which means you generally get compatible, or sometimes better performance compared to x86 and then the third reason why why arm became really popular is because it you can license anyone can license um IP, which means you can get really customizable with different things that you can do so uh on the left. That's the BBC micro, um the very first arm V1 arm machine. We have the iPhone Apple, creates the iPhone of arm V6. That's the Raspberry Pi.

A

A couple of years ago, AWS launched graviton, provide Army the cloud and then Apple also now provide you know, developer laptops that are based on M1 and M2 they're based on arm and then the other thing that makes arm quite different is arm, is really focused on the ecosystem and doesn't manufacture any kind of chips or do any manufacturing itself.

A

So, for Mom's point of view, the world looks like this: you have the Silicon ecosystem, which is what arm does has lots of Engineers who focus on designing architectures and what's coming next, uh it's really small. So you can't see this, but remember this Eda at the bottom right hand corner. It will come up again later. That's electronic design, automation. It will come up again above that you have the compute subsystem. So that's where you turn it into a CPU.

A

That's where I'm partners with lots of companies that manufacture the actual CPUs and then above that, that's where I play the software ecosystem. There's all the other things around it all the tools, all the libraries all the middleware, that you need to actually run.

A

So arm is now in all the major clouds. Apart from AWS, um you know: Google Cloud, Microsoft, Azure, the Chinese, the big ones, Baidu 10 cents, Alibaba and so on, and actually arm really is everywhere. This is a very incomplete list, but gives you a little bit of an idea of some of the open source, stuff open and closed Source stuff that supports um arms, also in cncf in LF, networking and a bunch of other Open Standards bodies.

A

Okay so multi-arch, so the goal of multi-arch infrastructure is that workloads can run on the best hardware for their price performance needs without developers being concerned about the underlying architecture. That's what it boils down to usually means moving from x86 to arm, but the ultimate goal of multi-arch is just use. Whatever is gives me the best price performance.

A

um I pulled these these headlines today this morning, just to give you an idea of what this actually looks like in practice. So most of the companies that are trying are trying to reduce costs effectively.

A

uh So there's- and these are some of the different companies that have confirmed they do use graviton, um quoting anywhere from what 25 up to 63 lower costs uh for moving to graviton.

A

And, of course you have to dog food, your own stuff, so remember, I, mentioned Eda. um That is part of the chip design process for arm very compute. Intensive workloads used to run on-prem x86 and um decided to move that to graviton.

A

And the results that they found was 60, better 60, better performance, half the cost and then saved a bunch of power as well.

A

Okay. This is my one slide where I'm gonna mention kubernetes, because I said this is the end of the day, so your brain's, all very tired already, but I'll give you some hints and tips. If you think this is something that you want to try one is kubernetes is not just kubernetes.

A

You have Prometheus and Falco and psyllium and flux or Argo or whatever. You have many many many applications that you run within kubernetes if you want to run production grade cluster.

A

So when we talk about oh moving on to kubernetes on arm, you actually have to look at many many different applications.

A

So your strategy for this could be you, move the control, plane nodes and leave the worker notes where they are. You could move the worker nodes, leave the control plane as it is. You could try and do a mix of x86 and arm worker nodes. Each of these has its own complications saves a certain amount adds a little bit of complexity in other ways when you're creating your clusters that, if you're, if that probably means you have to change how your clusters are created, that also causes problems.

A

If you're running things in Demon sets or thought causes problems, but you have to have multi-arged images that run because the Deep a demon set will run on every node.

A

So you have to make sure you're selecting the correct image for that architecture uh in order to select those that means that you need, you might have different limits and the requests per architecture um you need to control it using node affinity and teens and tolerations, and you think about whether you're building, if like, if you're, now getting to the point where you're like now I have to move my actual applications, then you know: are your bills reproducible? What do your pipelines look like? So none of this is simple.

A

I would be like having just told you you're like hey, you can save a bunch of money and save the planet. um This is like a year plus process to actually move it.

A

That's why it's hard, but at the same time that's kind of where the pressures are going now um and it's perhaps not uh uh it I think it's still pretty it's still kind of early now, but we're getting to this point where we're crossing the chasm with arm lots of companies are starting to do this migration because effectively all of this is a one-off cost, and then hopefully you continue to save 20 to 40 on your ongoing Cloud bills, forever more.

A

Okay, so takeaways takeaways. The main reason to move to arm is just cost savings you can get. You can see. We get between 20 and 40. Cheaper than running on x86 arm is causing the chasm um 48 out of the top 50 ec2 customers use graviton for some of their workloads, so it is real um whether you're at an end user company, that's trying to save money or whether you're a product company where your customers are starting to ask for arm it is coming it's coming in.

A

You should be aware about it and then, thirdly, just again people in timing, you know a lot of the stuff that we a lot of the reasons that we're here in these conferences is to talk to the people and to find out what's coming up next, so I really encourage you to just get together. Chat have fun, find out what's happening in the community, and that is it for me. Thank you so much these slides are on my blogshell.com. Thank you.

B

Thank you very much Cheryl. uh That was uh fantastically on time.

A

um I'm, like everyone, wants to get up and go yeah go for dinner.

C

I think so well Frozen, so they can't actually uh does anyone have questions for Cheryl.

B

Yes, we do have a question over there.

A

I see two there yeah.

C

Just a minor comment uh like I'm, really interested in adopting like our architecture and stuff but I, see in kubernetes space as a kind of major blocker is that um many people use those ugly charts made by Vietnamese and they read back all the images into like it's 86 air architecture, so I think, like Avengers, like arm, really should work with like those roadblocks to have avoid their adoption, because in the end, if someone like looks at the chart tries to deployed they just work and then all right any time on that so I'll see like like huge vendors.

C

They have a kind of capacity Workforce many connections, so they can like kind of convince.

C

Vietnamese to to do that, mentorship.

A

So you said they have ugly charts, you mean Helm, charts.

C

Yeah, it's like a huge repository of charts and they always repack uh like binaries into their own images based on living and it's never compatible with our yeah.

A

Yeah yeah. Well, luckily, my job well, my job title is the senior director of ecosystem. So it's actually my job to go out there and make it more beautiful. Make these companies I mean my goal really is to make the experiences smooth and easy as possible.

A

So I really want to work with these companies and, if anyone's here, actually you know if their company wants to work with arm and wants them wants to chat to me about it, like I, really encourage that, because yeah I agree a lot of it's really really ugly right now, and it could be a lot better.

B

A

B

um What I liked the most was choosing the best cost performance for your application? Do you see a web assembly as part of the motor Arc story say.

A

The last bit again, do you see.

B

Sorry, do you see webassembly as part of simultaneous architecture story, webassembly.

A

I, don't know where assembly in a lot of detail but I think from what I've seen it's taking off slowly so I'm sort of mildly, positive towards webassembly, but I'm not super familiar with it. Myself.

A

Cool anyone else thank.

C

A