Enterprises increasingly run the applications supporting their business processes in the cloud. Application data residing in the cloud expand the importance of cloud-based analytical workloads, which require provisioned infrastructure before any query processing can begin. Resource provisioning can be difficult for these workloads because they are often unpredictable and ad-hoc in nature. Overprovisioning and reduced cost-efficiency are the norm to avoid disruption of performance due to insufficient resources.
Recently, cloud providers have introduced means to allocate and bill fine-granular units of resources with function-as-a-service (FaaS) compute platforms and shared object storage systems. We evaluate this so-called serverless infrastructure regarding its performance elasticity and variability. Based on our findings, we build the Skyrise serverless query processor that interhits the elastic scalability of its underlying FaaS infrastructure while it deals with the limitations and inefficiencies. Skyrise enables cost-efficient, interactive analytics on infrequently accessed data, a workload for which conventionally provisioned database systems are idle most of the time and thus not viable.