SanssouciDB: an in-memory database for processing enterprise workloads

Abstract

In this paper, we present SanssouciDB: a database system designed for serving ERP transactions and analytics out of the same data store. It consists of a columnstore engine for high-speed analytics and transactions on sparse tables, as well as an engine for so-called combined columns, i.e., column groups which are used for materializing result sets, intermediates, and for pocessing transactions on tables touching many attributes at the same time. Our analysis of SAP customer data showed that the vast majority of transactions in an ERP system are of analytical nature. We describe the key concepts of SanssouciDB's architecture: concurrency control, techniques for compression and parallelization, and logging. To illustrate the potential of combining OLTP and OLAP processing in the same database, we give several examples of new applications which have been built on top of an early version of SanssouciDB and discuss the speedup achieved when running these applications at SAP customer sites.