11-10, 15:10–15:40 (MST), Flex Space
If you’ve ever asked yourself the question, “I’m running Cluster Autoscaler right now, should I switch to Karpenter?”, then this talk is for you! These two projects are alternative solutions for autoscaling nodes in a Kubernetes cluster based on demand. While they share many similarities, there are some important differences as well–differences that can have a meaningful impact on the cost and reliability of your infrastructure. Join us for a data-driven review of the features and limitations of each option. Attendees will learn how workload configuration and cloud inventory affect the decision making process. By examining data from controlled experiments in simulated and live environments, the presenters will illuminate important areas of focus to evaluate when making your choice. At the end of this presentation, you will be equipped with data and tools needed to discover which autoscaler is best for your use case.
Audience members will come away from this presentation with two specific takeaways: first will be a better understanding of Cluster Autoscaler and Karpenter, and situations in which one alternative might be better than the other. Second will be an understanding of tools and methods for performing data analysis on Kubernetes cluster configuration and operation.
The cloud-native ecosystem can be confusing to navigate; many products operate in similar spaces, and it can be difficult to understand what the “right” choice is for an organization. Using Karpenter and Cluster Autoscaler as a motivating example, we will show cluster operators how to answer questions like, “Which one’s faster? Which one makes better placement decisions? Which one’s cheaper?” While the comparison between Cluster Autoscaler and Karpenter is timely, given Karpenter’s acceptance into SIG-Autoscaling last year, we hope that this talk can show cluster operators how to make informed, data-driven decisions about all areas of the infrastructure they manage, not just autoscaling.
David Morrison is a research scientist at Applied Computing Research Labs, an open-source research and development lab exploring scheduling and optimization problems in distributed computing. Previously, David was a staff engineer at Airbnb and at Yelp. David received his PhD in Computer Science from the University of Illinois, Urbana-Champaign in 2014. He is a prolific public speaker, having given presentations at KubeCon, LISA, MesosCon, and others.
Connect on Mastodon! https://hachyderm.io/@drmorr
Michael McCune is a software developer creating open source infrastructure and applications for cloud platforms. He has a passion for problem solving and team building, and a lifelong love of music, food, and culture.