Distributed SQL Summit 2021

Spot instances offer upto 90% cost savings over reserved instances. This session explores the feasibility of using spot instances to lower the cost of running a YugaybteDB cluster for workloads that require performance and resilience. Customers can get high throughput reads and writes at a fraction of the cost while maintaining complete stability even when spot instances are taken away without notice. Cast handles the interruptions seamlessly and replaces capacity. YugabyteDB is tolerant to these interruptions, and is able to self heal and replicate even when VM’s are arbitrarily removed.


During the session, we will deploy a 5-node YugabyteDB cluster on a single cloud in a managed Kubernetes cluster, with 2 nodes being on-demand (regular pricing) and 3 nodes being spot (heavily discounted pricing). We will assess the impact on performance and resilience from using ephemeral spot instances to run a database. We will show a running benchmark while interruptions are happening.