Paper Title: Running Stafeful Workloads on Flash and Kubernetes: Challenges and Solutions

Paper Abstract: Running stateful workloads such as Databases and Big Data on Kubernetes is hard. Storage stacks need to support more than the basic ability to provision storage volumes to Pods inside which stateful workloads run. Stateful workloads are performance-sensitive, requiring the correct binding between apps and flash storage to ensure predictable performance and higher utilization in a multi-tenant consolidated Kubernetes environment. The design of Kubernetes makes storage and data management very difficult for deploying stateful workloads such as Hadoop, Oracle, Splunk, Mongo, Elastic, SAP HANA, etc. This presentation covers both the challenges one needs to overcome at the Storage layer, as well as solutions to enable both deployment and lifecycle management of these workloads on a modern platform like Kubernetes. It also covers how stateful applications can leverage the Kubernetes Operator framework to better consume flash storage based on the performance SLAs of the workload.

Paper Author: Jagadish Mukku, Sr Architect, Robin.io

Author Bio: Jagadish Mukku has more than 18 years of product and technology experience with extensive experience in storage. Currently at robin.io, he architected, designed and implemented data management for stateful applications in kubernetes. In previous positions he worked at violin systems as a software architect on data management software(deduplication, snapshots, thin-provisioning, replication) for an AFA(All Flash Array), worked at veritas and agami systems(NAS appliance startup). He holds a master’s degree in computational science from Indian Institute of Science (IISc), Bangalore.