摘要：
The hardware landscape for database systems has changed dramatically over the past two decades. Today, the traditional database system architecture that was pio- neered by System R and is implemented by IBM DB2, Microsoft SQL Server, Oracle and Postgres, shows weaknesses in all major areas of application of commercial da- tabase systems: operational transaction processing, decision support and business intelligence, and beyond relational workloads. The shortcomings of the traditional architecture on modern hardware have led to the development of new data man- agement approaches and overthrew common beliefs in how to build a database system. However, most modern database systems are optimized for only one of the aforementioned areas of application. This separation into specialized systems makes it difficult to achieve real-time awareness and to get to a common under- standing of all data across an organization. HyPer is a modern hybrid high performance main-memory database system that overcomes the weaknesses of the traditional architecture and aims at fulfilling the vision of a one size fits all database management system. Using a novel and unique architecture, HyPer is able to unify the capabilities of modern specialized transac- tional and analytical systems in a single system without sacrificing performance or standards and reliability guarantees. In addition, HyPer aspires to integrate be- yond relational workloads to overcome the connection gap between the relational and beyond relational world. This thesis makes several contributions to the research area of main-memory da- tabase systems and the HyPer main-memory database system by improving the scalability and flexibility of query and transaction processing. In particular, we contribute (i) a fast serializable multi-version concurrency mechanism, (ii) an ap- proach for fast data ingestion and in-situ query processing on files, ( ) a scale-out of the HyPer system that allows elastic query processing on transac
