Episode #85: Serverless at IBM with Michael Behrendt

Serverless Chats

About Michael Behrendt

Michael Behrendt is a Distinguished Engineer in the IBM Cloud development organization. He is responsible for IBM’s technical strategy for offerings around serverless & Function-as-a-Service. Before that, he was the Chief Architect of IBM's core cloud platform and was one of the initial founding members incubating it, led the development of IBM's Cloud Computing Reference Architecture, was a worldwide field-facing cloud architect for many years, and drove key product incubation & development activities for IBM's cloud portfolio Michael has been working on Cloud Computing for more than 15 years and has 37 patents. He is located in the IBM Research & Development Laboratory in Boeblingen, Germany.

• Twitter: @Micheal_BEH
• LinkedIn: Michael Behrendt
• IBM Cloud Code Engine: https://ibm.biz/codeengine
• Cloud Functions: https://cloud.ibm.com/functions/
• IBM Cloud Functions Tutorial: https://cloud.ibm.com/functions/
• IBM Cloud Code Engine Getting Started: https://cloud.ibm.com/docs/codeengine?topic=codeengine-getting-started
• IBM Cloud Free Tier: https://www.ibm.com/cloud/free

Watch this episode on YouTube: https://youtu.be/t3KHoCAVazU

Transcript

Jeremy: Hi everyone. I'm Jeremy Daly, and this is Serverless Chats. Today I'm speaking with Michael Behrendt. Hey Michael. Thanks for joining me.

Michael: Hey, Jeremy. Thanks for having me.

Jeremy: So you are a distinguished engineer, chief architect, Serverless IBM Cloud at IBM. So why don't you tell the listeners a little bit about your background and what you do at IBM?

Michael: Sure. Thank you. So I've been working at IBM in various technical roles over the last 15 to 20 years. I have been in product development, product incubation, I've been working in the field as a workload architect. And for the last 10 years as well I've been working in the Cloud division in itself, working on various topics, incubating it and so on. And since about six years now, I'm really focused on serverless as a topic as a whole. So that's what I'm doing most of my time. Working with customers, working on product development, making architectural decisions, technology decisions, and so on.

Jeremy: Awesome. All right. First of all, I want to thank IBM for sponsoring this episode. So that's great continuing to support the community and continuing to invest in serverless. And when it comes to serverless at IBM, you are the guy. You were there right back in the beginning. I had Rodric Rabbah on the show a couple of weeks ago. And we were talking about how it all got started. But I know you have a bunch of stories as well. So what if we go all the way back and start that sort of six years ago and talk about how did it begin? How did serverless at IBM sort of get kicked off?

Michael: There is some interesting stories there. So long a while ago now, I've been looking into the serverless market as it was evolving, what was happening in the field, what customers are doing. And I felt like we need to do something in the serverless space as well. And by purpose, I thought we shouldn't be starting this as a right off the bat product development effort, but rather since it was such a new space do some exploratory stuff first and have it really open-ended in terms of what we are going to end up with from a technology perspective.

So I was in Beijing for a business trip and I had a call with a VP for research at IBM for Cloud. And I still remember it was 10:00 PM at night. And we talked about we need to do something in that space. So we agreed on that call, let's do something in that space. And he basically then brought in a team from the research side, Rodric was part of the team to kick off that whole effort.

Jeremy: Right. So I don't think I've ever heard a story that starts 10:00 PM in Beijing, ever heard a story that didn't end, or it didn't have an exciting ending to it. So all right. So you brought in this team to kind of start working on it. And so what did you do first? What was the initial goal? I mean, you were surveying the market, doing the research, as you said. So sort of, how did you sort of take those first steps?

Michael: So we put together this team of really talented people in research, and we basically set up our goal. It's what do we want to accomplish from a workload perspective? Which kind of workloads do we want to support? We want to allow composition of functions, something we are talking about these days as well, but it was like a new concept back then. We wanted to be able to be very flexible in terms of which kinds of workloads people can run. Should it only be functions or should it be more cost in the workloads as well? So we went into different directions.

We looked at non-functionals like, how quickly should it be possible to deploy a new function or update a function to have a very quick interloop development cycle. And that drove lots of technology and design decisions. And we've been running that with playbacks every week I believe, where the team played back to a broader group of people like what they were doing, their findings and so on. And we iterate it all way towards into that. And one of the big milestones that was at the end of this first wave was OpenWhisk, as an open source project.

Jeremy: Right. So what were some of those early use cases? Because that was one of the things when serverless sort of first started coming out. And again, OpenWhisk is a fast, functions as a service similar to Lambda or a Google Cloud Functions, things like that. But what were those early use cases? Because I remember way back in the beginning, it was very, very limited.

Michael: Yeah. So I think one of the first use cases was, and that is a bread and butter use case these days as well, still it's those HDP endpoints. That was a very broadly applicable horizontal in many industries applicable use case. Another one that I still remember the specific customers we were working with in these days was data processing, like objects or photos in particular that had to be processed in a certain way, like auto cropping, auto sharpening, object detection, storing metadata.

And I still remember we had one of our very first customers, they went GA while we were still in beta. And so because they felt good with what they had. And I still remember talking to the CEO one time and he said, in the early days, their operations guy talked to him and asked whether our billing engine was broken because the bill was so low. And they came in from a past background. So they moved from a past to function as a service and what they saw was 10X performance increase in combination with 90% cost reduction. And that was just astonishing to them which they had never seen before.

Jeremy: Right. Oh, that's amazing. So we can't talk about functions as a service without sort of talking about the 800 pound gorilla in the room, which is AWS Lambda. But I know that, and this is something I actually really appreciate about what's happening in the serverless mo...