Driven by research and adopted by the industry, two major IT domains have experienced a tremendous increase in service offers over the last few years: database management systems (DBMS) and cloud computing. Since the one-size-fits-all paradigm is no longer valid, the relational DBMS landscape has been extended with over 250 (and still growing) of NoSQL and NewSQL DBMS that promise to provide the non-functional features high performance, scalability, elasticity, and high availability for any data-intensive use case. On the resource level, cloud resources have become the preferred option to operate modern data-intensive applications in order to enable “unlimited scalability” and elasticity on the resource level. In consequence, the cloud market has become heterogeneous with over 20,000 public cloud resource offers. Needless to say, cloud resources are also a common way to operate DBMS, especially as distributed NoSQL and NewSQL DBMS promise to be cloud-native. Moreover, many established DBMS providers extend their service portfolio with Database-as-a-Service (DBaaS). While these DBMS and cloud advancements enable tailor-made data storage solutions, finding these solutions is a challenging process that involves in-depth benchmarking of the relevant non-functional features. However, reliable benchmarking of cloud-hosted DBMS requires multi-domain knowledge, reproducibility, and multi-level result processing. In this talk, we report on our long-term experiences in benchmarking cloud-hosted DBMS by highlighting key impact factors for significant evaluations. In addition, we discuss relevant DBMS performance, scalability, elasticity, availability, and cost metrics. These concepts are integrated into our benchmarking-as-a-service platform and will be demonstrated by a set of real-world cases, addressing challenges such as: Which cloud provider does provide the best DBMS performance/cloud cost ratio for my application? Will I get more performance with the next DBMS version? From a performance perspective, which one to choose: self-hosted DBMS vs. DBaaS?
- Assess the key impact factors for benchmarking cloud-hosted DBMS
- Specify comprehensive evaluations for cloud resources and distributed DBMS
- Measure the higher-level DBMS metrics scalability, elasticity, and availability
Daniel Seybold is a researcher in the area of cloud computing with a focus on distributed databases in the cloud. Further interests cover cloud orchestration, model-driven engineering, and performance evaluations of distributed systems. Daniel is currently Chief Technical Officer of benchANT.