RTP: robust tenant placement for elastic in-memory database clusters

Abstract

In the cloud services industry, a key issue for cloud operators is to minimize operational costs. In this paper, we consider algorithms that elastically contract and expand a cluster of in-memory databases depending on tenants' behavior over time while maintaining response time guarantees. We evaluate our tenant placement algorithms using traces obtained from one of SAP's production on-demand applications. Our experiments reveal that our approach lowers operating costs for the database cluster of this application by a factor of 2.2 to 10, measured in Amazon EC2 hourly rates, in comparison to the state of the art. In addition, we carefully study the trade-off between cost savings obtained by continuously migrating tenants and the robustness of servers towards load spikes and failures.