Skip to content

High Performance Kubernetes Platform For Stateful Workloads

Summary

Enterprises need high-performance storage for their stateful applications. Their needs vary from running Relational and NoSQL Databases to Big Data to AI/ML workloads. They want near bare metal like performance along with scale. In this white paper, we compare Robin Platform with open source GlusterFS and our analysis show that Robin platform provides near bare-metal performance and could scale well to meet the needs of Big Data and AI/ML applications.

Introduction

As containers gain traction inside the enterprise and Kubernetes emerges as the standard for container orchestration, users are looking to leverage the technology for running mission-critical workloads such as stateful applications like databases, big data and AI/ML applications. Unlike stateless applications, these applications have important storage and networking requirements. The containers and Kubernetes are helping modern enterprises to embrace agile and DevOps while also bringing better efficiencies to IT operations. The key to enterprise IT transformation lies in leveraging a stack such as Kubernetes for both states, stateless and stateful, for big data applications and AI/ML.

The Kubernetes community has focused their attention on the need to support stateful workloads – the work done around StatefulSets is a good indicator of the progress. But this effort is far from mature and there exists operational overhead in provisioning the clusters needed for persistent volumes. Many IT organizations are spending multiple cycles to get Kubernetes set up for stateful workloads, leading to friction and delays. The problem gets bigger when big data and other data-intensive workloads become part of the equation. Beyond the operational overhead, performance is also a critical criterion for these workloads. The enterprise decision makers are torn between selecting a DIY approach to running stateful workloads on Kubernetes and finding the right platform that
is suitable for data-intensive workloads.

If faster time to market is the driver for the adoption of cloud and IT modernization strategies, building a platform for stateful workloads from vanilla Kubernetes or one of the platforms available for stateless applications is a waste of resources leading to undifferentiated heavy lifting. The key to successful digital transformation lies in picking a platform that provides a competitive edge in the market without incurring high operational overhead.

Key evaluation criteria

– Does the Kubernetes platform give bare metal like performance?
– Are the performance guarantees available at scale?
– Is the performance predictable?