Orchestrating DBMS Benchmarking in the Cloud with Kubernetes

Containerization has become a common practise in software provisioning. Kubernetes (K8s) is useful in deploying containers in clusters, in managing their lifecycle, in scheduling and resource allocation. The benchmarking process requires the interaction of various components. We propose a way to organize benchmarking in the Cloud by looking at typical components in the process and ask if they could be managed by K8s as containerized Microservices. We aim at scalability for the process, parallelized execution and minimized traffic I/O from and into the Cloud. This supports planning a series of experiments to investigate a high-dimensional parameter space and avoiding complex installations. This also provides a way for Cross-Cloud comparison. In this article we discuss 1. how objects of K8s can match components of a benchmarking process, 2. how to orchestrate the benchmarking workflow in K8s. We also present an implementation. We show this approach is feasible, relevant, portable and scalable by designing and inspecting a basic profiling benchmark on TPC-DS data handled by 13 DBMS at two private Clouds and five commercial Cloud providers.