Computing hardware is changing from systems with homogeneous CPUs to systems with heterogeneous computing units like GPUs, Many Integrated Cores, or FPGAs. This trend is caused by scaling problems of homogeneous systems, where heat dissipation and energy consumption is limiting further growths in compute-performance. Heterogeneous systems provide differently optimized computing hardware, which allows different operations to be computed on the most appropriate computing unit, resulting in faster execution and less energy consumption. For database systems, this is a new opportunity to accelerate query processing, allowing faster and more interactive querying of large amounts of data. However, the current hardware trend is also a challenge as most database systems do not support heterogeneous computing resources and it is not clear how to support these systems best. In the past, mainly single operators were ported to different computing units showing great results, while missing a system wide application. To efficiently support heterogeneous systems, a systems approach for query processing and query optimization is needed. In this thesis, we tackle the optimization challenge in detail. As a starting point, we evaluate three different approaches on isolated use-cases to assess their advantages and limitations. First, we evaluate a fork-join approach of intra-operator parallelism, where the same operator is executed on multiple computing units at the same time, each execution with different data partitions. Second, we evaluate using one computing unit statically to accelerate one operator, which provides high code-optimization potential, due to this static and pre-known usage of hardware and software. Third, we evaluate dynamically placing operators onto computing units, depending on the operator, the available computing hardware, and the given data sizes. We argue that the first and second approach suffer from multiple overheads or high implementation costs. The third approach, dynamic placement, shows good performance, while being highly extensible to different computing units and different operator implementations. To automate this dynamic approach, we first propose general placement optimization for query processing. This general approach includes runtime estimation of operators on different computing units as well as two approaches for defining the actual operator placement according to the estimated runtimes. The two placement approaches are local optimization, which decides the placement locally at run-time, and global optimization, where the placement is decided at compile-time, while allowing a global view for enhanced data sharing. The main limitation of the latter is the high dependency on cardinality estimation of intermediate results, as estimation errors for the cardinalities propagate to the operator runtime estimation and placement optimization. Therefore, we propose adaptive placement optimization, allowing the placement optimization to become fully independent of cardinalities estimation, effectively eliminating the main source of inaccuracy for runtime estimation and placement optimization. Finally, we define an adaptive placement sequence, incorporating all our proposed techniques of placement optimization. We implement this sequence as a virtualization layer between the database system and the heterogeneous hardware. Our implementation approach bases on preexisting interfaces to the database system and the hardware, allowing non-intrusive integration into existing database systems. We evaluate our techniques using two different database systems and two different OLAP benchmarks, accelerating the query processing through heterogeneous execution.

Computing hardware is changing from systems with homogeneous CPUs to systems with heterogeneous computing units like GPUs, Many Integrated Cores, or FPGAs. This trend is caused by scaling problems of homogeneous systems, where heat dissipation and energy consumption is limiting further growths in compute-performance. Heterogeneous systems provide differently optimized computing hardware, which allows different operations to be computed on the most appropriate computing unit, resulting in faster execution and less energy consumption. For database systems, this is a new opportunity to accelerate query processing, allowing faster and more interactive querying of large amounts of data. However, the current hardware trend is also a challenge as most database systems do not support heterogeneous computing resources and it is not clear how to support these systems best. In the past, mainly single operators were ported to different computing units showing great results, while missing a system wide application. To efficiently support heterogeneous systems, a systems approach for query processing and query optimization is needed. In this thesis, we tackle the optimization challenge in detail. As a starting point, we evaluate three different approaches on isolated use-cases to assess their advantages and limitations. First, we evaluate a fork-join approach of intra-operator parallelism, where the same operator is executed on multiple computing units at the same time, each execution with different data partitions. Second, we evaluate using one computing unit statically to accelerate one operator, which provides high code-optimization potential, due to this static and pre-known usage of hardware and software. Third, we evaluate dynamically placing operators onto computing units, depending on the operator, the available computing hardware, and the given data sizes. We argue that the first and second approach suffer from multiple overheads or high implementation costs. The third approach, dynamic placement, shows good performance, while being highly extensible to different computing units and different operator implementations. To automate this dynamic approach, we first propose general placement optimization for query processing. This general approach includes runtime estimation of operators on different computing units as well as two approaches for defining the actual operator placement according to the estimated runtimes. The two placement approaches are local optimization, which decides the placement locally at run-time, and global optimization, where the placement is decided at compile-time, while allowing a global view for enhanced data sharing. The main limitation of the latter is the high dependency on cardinality estimation of intermediate results, as estimation errors for the cardinalities propagate to the operator runtime estimation and placement optimization. Therefore, we propose adaptive placement optimization, allowing the placement optimization to become fully independent of cardinalities estimation, effectively eliminating the main source of inaccuracy for runtime estimation and placement optimization. Finally, we define an adaptive placement sequence, incorporating all our proposed techniques of placement optimization. We implement this sequence as a virtualization layer between the database system and the heterogeneous hardware. Our implementation approach bases on preexisting interfaces to the database system and the hardware, allowing non-intrusive integration into existing database systems. We evaluate our techniques using two different database systems and two different OLAP benchmarks, accelerating the query processing through heterogeneous execution.

Nowadays, increasingly large amounts of data are being collected in numerous areas ranging from science to industry. To gain valueable insights from these data, the importance of Online Analytical Processing (OLAP) workloads is constantly growing. At the same time, the hardware landscape is continuously evolving. On the one hand, the increasing capacities of DRAM allow database systems to store their entire data in main memory. Furthermore, the performance of microprocessors has improved tremendously in recent years through the use of sophisticated hardware techniques, such as Single Instruction Multiple Data (SIMD) extensions promising hitherto unknown processing speeds. On the other hand, the main memory bandwidth has not increased proportionately, such that the data access is now the main bottleneck for an efficient data processing. To face these developments, in-memory column-stores have emerged as a new database architecture. These systems store each attribute of a relation separately in memory as a contiguous sequence of values. It is state-of-the-art to encode all values as integers and apply lossless lightweight integer compression to reduce the data size. This offers several advantages ranging from lower transfer times between RAM and CPU over a better utilization of the cache hierarchy to fast direct processing of compressed data. However, compression also incurs a certain computational overhead. State-of-the-art systems focus on the compression of base data. However, intermediate results generated during the execution of complex analytical queries can exceed the base data in number and total size. Since in in-memory systems, accessing intermediates is as expensive as accessing base data, intermediates should be handled as efficiently as possible, too. While there are approaches trying to avoid intermediates whenever it is possible, we envision the orthogonal approach of efficiently representing intermediates using lightweight integer compression algorithms to reduce memory accesses. More precisely, our vision is a balanced query processing based on lightweight compression of intermediate results in in-memory column-stores. That means, all intermediates shall be represented using a suitable lightweight integer compression algorithm and processed by compression-enabled query operators to avoid a full decompression, whereby compression shall be used in a balanced way to ensure that its benefits outweigh its costs. In this thesis, we address all important aspects of this vision. We provide an extensive overview of existing lightweight integer compression algorithms and conduct a systematical experimental survey of several of these algorithms to gain a deep understanding of their behavior. We propose a novel compression-enabled processing model for in-memory column-stores allowing a continuous compression of intermediates. Additionally, we develop novel cost-based strategies for a compression-aware secondary query optimization to make effective use of our processing model. Our end-to-end evaluation using the famous Star Schema Benchmark shows that our envisioned compression of intermediates can improve both the memory footprint and the runtime of complex analytical queries significantly.:1 Introduction 1.1 Contributions 1.2 Outline 2 Lightweight Integer Compression 2.1 Foundations 2.1.1 Disambiguation of Lightweight Integer Compression 2.1.2 Overview of Lightweight Integer Compression 2.1.3 State-of-the-Art in Lightweight Integer Compression 2.2 Experimental Survey 2.2.1 Related Work 2.2.2 Experimental Setup and Methodology 2.2.3 Evaluation of the Impact of the Data Characteristics 2.2.4 Evaluation of the Impact of the Hardware Characteristics 2.2.5 Evaluation of the Impact of the SIMD Extension 2.3 Summary and Discussion 3 Processing Compressed Intermediates 3.1 Processing Model for Compressed Intermediates 3.1.1 Related Work 3.1.2 Description of the Underlying Processing Model 3.1.3 Integration of Compression into Query Operators 3.1.4 Integration of Compression into the Overall Query Execution 3.1.5 Efficient Implementation 3.1.6 Evaluation 3.2 Direct Integer Morphing Algorithms 3.2.1 Related Work 3.2.2 Integer Morphing Algorithms 3.2.3 Example Algorithms 3.2.4 Evaluation 3.3 Summary and Discussion 4 Compression-Aware Query Optimization Strategies 4.1 Related Work 4.2 Compression-Aware Secondary Query Optimization 4.2.1 Compression-Level: Selecting a Suitable Algorithm 4.2.2 Operator-Level: Selecting Suitable Input/Output Formats 4.2.3 QEP-Level: Selecting Suitable Formats for All Involved Columns 4.3 Evaluation 4.3.1 Compression-Level: Selecting a Suitable Algorithm 4.3.2 Operator-Level: Selecting Suitable Input/Output Formats 4.3.3 Lessons Learned 4.4 Summary and Discussion 5 End-to-End Evaluation 5.1 Experimental Setup and Methodology 5.2 A Simple OLAP Query 5.3 Complex OLAP Queries: The Star Schema Benchmark 5.4 Summary and Discussion 6 Conclusion 6.1 Summary of this Thesis 6.2 Directions for Future Work Bibliography List of Figures List of Tables