Building Scalable and Robust ML Pipelines and Storage with Kubeflow

This forward-looking presentation and demonstration will focus on the data storage requirements of stateful Kubernetes applications, especially for machine learning use cases that leverage Kubeflow to deliver Tensorflow models. It will review the common configurations to support stateful Kubernetes applications, along with the benefits and challenges of each option.

The presentation will detail the new 2020 Kubernetes Stateful Application Architectures, which leverage standards-based interfaces, secondary storage functionality, and local disks to deliver significantly improved storage performance and lower costs. It will review a popular Kubeflow Pipelines use case and the delivery of efficient versioning, packaging, reproducibility, and compliance. It will also explain the machine learning requirements for advanced data management solutions, which simplify portability, improve performance, and minimize network traffic and storage requirements.

About the Presenter
Josh Bottum supports the Kubeflow Community as a member of the Product Management Team. He is also a Vice President at Arrikto (https://www.arrikto.com/). Arrikto is a San Mateo based start-up that develops standards-based solutions for stateful Kubernetes applications. Arrikto is a code contributor to several of the Kubeflow Working Groups and supports the development of Storage, Notebooks, Pipelines, Laptop, and On-prem functionalities.