Querying and efficiently analyzing complex data is required to gain valuable business insights, to support machine learning applications, and to make up-to-date information available. Therefore, this thesis investigates opportunities and challenges of selecting the most efficient execution strategy for analytical queries. These challenges include hard-to-capture data characteristics such as skew and correlation, the support of arbitrary data types, and the optimization time overhead of complex queries. Existing approaches often rely on optimistic assumptions about the data distribution, which can result in significant response time delays when these assumptions are not met. On the contrary, we focus on robust query optimization, emphasizing consistent query performance and applicability. Our presentation follows the general select-project-join query pattern, representing the fundamental stages of analytical query processing. To support arbitrary data types and complex filter expressions in the select stage, a novel sampling-based selectivity estimator is developed. Our approach exploits information from filter subexpressions and estimates correlations that are not captured by existing sampling-based methods. We demonstrate improved estimation accuracy and query execution time. Further, to minimize the runtime overhead of sampling, we propose new techniques that exploit access patterns and auxiliary database objects such as indices. For the join stage, we introduce a robust optimization approach by developing an upper-bound join enumeration strategy that connects accurate filter selectivity estimates –e.g., using our sampling-based approach– to join ordering. We demonstrate that join orders based on our upper-bound join ordering strategy achieve more consistent performance and faster workload execution on state-of-the-art database systems. However, besides identifying good logical join orders, it is crucial to determine appropriate physical join operators before query plan execution. To understand the importance of fine-grained physical operator selections, we exhaustively execute fixed join orders with all possible operator combinations. This analysis reveals that none of the investigated query optimizers fully reaches the potential of optimal operator decisions. Based on these insights and to achieve fine-grained operator selections for the previously determined join orders, the thesis presents a lightweight learning-based physical execution plan refinement component called. We show that this refinement component consistently outperforms existing approaches for physical operator selection while enabling a novel two-stage optimizer design. We conclude the thesis by providing a framework for the two-stage optimizer design that allows users to modify, replicate, and further analyze the concepts discussed throughout this thesis.:1 INTRODUCTION 1.1 Analytical Query Processing . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2 Select-Project-Join Queries . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.3 Basics of SPJ Query Optimization . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.1 Plan Enumeration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 Cost Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.3.3 Cardinality Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.4 Robust SPJ Query Optimization . . . . . . . . . . . . . . . . . . . . . . . . 16 1.4.1 Tail Latency Root Cause Analysis . . . . . . . . . . . . . . . . . . . 17 1.4.2 Tenets of Robust Query Optimization . . . . . . . . . . . . . . . . . 19 1.5 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.6 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2 SELECT (-PROJECT) STAGE 2.1 Sampling for Selectivity Estimation . . . . . . . . . . . . . . . . . . . . . . 24 2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.2.1 Combined Selectivity Estimation (CSE) . . . . . . . . . . . . . . . . 29 2.2.2 Kernel Density Estimator . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.2.3 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.3 Beta Estimator for 0-Tuple-Situations . . . . . . . . . . . . . . . . . . . . . 33 2.3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.3.2 Beta Distribution in Non-0-TS . . . . . . . . . . . . . . . . . . . . . . 35 2.3.3 Parameter Estimation in 0-TS . . . . . . . . . . . . . . . . . . . . . . 37 2.3.4 Selectivity Estimation and Predicate Ordering . . . . . . . . . . . 39 2.3.5 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.4 Customized Sampling Techniques . . . . . . . . . . . . . . . . . . . . . . 53 2.4.1 Focused Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.4.2 Conditional Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . 56 2.4.3 Zone Pruning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 2.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3 JOIN STAGE: LOGICAL ENUMERATION 3.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.1.1 Point Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.1.2 Join Cardinality Upper Bound . . . . . . . . . . . . . . . . . . . . . 64 3.2 Upper Bound Join Enumeration with Synopsis (UES) . . . . . . . . . . . . 66 3.2.1 U-Block: Simple Upper Bound for Joins . . . . . . . . . . . . . . . . 67 3.2.2 E-Block: Customized Enumeration Scheme . . . . . . . . . . . . . 68 3.2.3 UES Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 3.3.1 General Performance . . . . . . . . . . . . . . . . . . . . . . . . . . 72 3.3.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4 JOIN STAGE: PHYSICAL OPERATOR SELECTION 4.1 Operator Selection vs Join Ordering . . . . . . . . . . . . . . . . . . . . . 77 4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4.2.1 Adaptive Query Processing . . . . . . . . . . . . . . . . . . . . . . 80 4.2.2 Bandit Optimizer (Bao) . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.3 TONIC: Learned Physical Join Operator Selection . . . . . . . . . . . . . 82 4.3.1 Query Execution Plan Synopsis (QEP-S) . . . . . . . . . . . . . . . 83 4.3.2 QEP-S Life-Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 4.3.3 QEP-S Design Considerations . . . . . . . . . . . . . . . . . . . . . . 87 4.4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.4.1 Performance Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 4.4.2 Rate of Improvement . . . . . . . . . . . . . . . . . . . . . . . . . . 92 4.4.3 Data Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.4.4 TONIC - Runtime Traits . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5 TWO-STAGE OPTIMIZER FRAMEWORK 5.1 Upper-Bound-Driven Join Ordering Component . . . . . . . . . . . . . 101 5.2 Physical Operator Selection Component . . . . . . . . . . . . . . . . . . 103 5.3 Example Query Optimization . . . . . . . . . . . . . . . . . . . . . . . . . 103 6 CONCLUSION 107 BIBLIOGRAPHY 109 LIST OF FIGURES 117 LIST OF TABLES 121 A APPENDIX A.1 Basics of Query Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 A.2 Why Q? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 A.3 0-TS Proof of Unbiased Estimate . . . . . . . . . . . . . . . . . . . . . . . . 125 A.4 UES Upper Bound Property . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 A.5 TONIC – Selectivity-Aware Branching . . . . . . . . . . . . . . . . . . . . . 128 A.6 TONIC – Sequences of Query Execution . . . . . . . . . . . . . . . . . . . 129

Good cardinality estimates are critical for generating good execution plans during query optimization. Complex predicates, correlations between columns, and user-defined functions are extremely hard to handle when using the traditional histogram approach. This demo illustrates the use of sample views for cardinality estimations as prototyped in Microsoft SQL Server. We show the creation of sample views, discuss how they are exploited during query optimization, and explain their potential effect on query plans. In addition, we also show our implementation of maintenance policies using statistical quality control techniques based on query feedback.

Querying and efficiently analyzing complex data is required to gain valuable business insights, to support machine learning applications, and to make up-to-date information available. Therefore, this thesis investigates opportunities and challenges of selecting the most efficient execution strategy for analytical queries. These challenges include hard-to-capture data characteristics such as skew and correlation, the support of arbitrary data types, and the optimization time overhead of complex queries. Existing approaches often rely on optimistic assumptions about the data distribution, which can result in significant response time delays when these assumptions are not met. On the contrary, we focus on robust query optimization, emphasizing consistent query performance and applicability. Our presentation follows the general select-project-join query pattern, representing the fundamental stages of analytical query processing. To support arbitrary data types and complex filter expressions in the select stage, a novel sampling-based selectivity estimator is developed. Our approach exploits information from filter subexpressions and estimates correlations that are not captured by existing sampling-based methods. We demonstrate improved estimation accuracy and query execution time. Further, to minimize the runtime overhead of sampling, we propose new techniques that exploit access patterns and auxiliary database objects such as indices. For the join stage, we introduce a robust optimization approach by developing an upper-bound join enumeration strategy that connects accurate filter selectivity estimates –e.g., using our sampling-based approach– to join ordering. We demonstrate that join orders based on our upper-bound join ordering strategy achieve more consistent performance and faster workload execution on state-of-the-art database systems. However, besides identifying good logical join orders, it is crucial to determine appropriate physical join operators before query plan execution. To understand the importance of fine-grained physical operator selections, we exhaustively execute fixed join orders with all possible operator combinations. This analysis reveals that none of the investigated query optimizers fully reaches the potential of optimal operator decisions. Based on these insights and to achieve fine-grained operator selections for the previously determined join orders, the thesis presents a lightweight learning-based physical execution plan refinement component called. We show that this refinement component consistently outperforms existing approaches for physical operator selection while enabling a novel two-stage optimizer design. We conclude the thesis by providing a framework for the two-stage optimizer design that allows users to modify, replicate, and further analyze the concepts discussed throughout this thesis.:1 INTRODUCTION 1.1 Analytical Query Processing . . . . . . . . . . . . . . . . . . . 12 1.2 Select-Project-Join Queries . . . . . . . . . . . . . . . . . . . 13 1.3 Basics of SPJ Query Optimization . . . . . . . . . . . . . . . . . 14 1.3.1 Plan Enumeration . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 Cost Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 1.3.3 Cardinality Estimation . . . . . . . . . . . . . . . . . . . . . 15 1.4 Robust SPJ Query Optimization . . . . . . . . . . . . . . . . . . 16 1.4.1 Tail Latency Root Cause Analysis . . . . . . . . . . . . . . . . 17 1.4.2 Tenets of Robust Query Optimization . . . . . . . . . . . . . . 19 1.5 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.6 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2 SELECT (-PROJECT) STAGE 2.1 Sampling for Selectivity Estimation . . . . . . . . . . . . . . . 24 2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.2.1 Combined Selectivity Estimation (CSE) . . . . . . . . . . . . . 29 2.2.2 Kernel Density Estimator . . . . . . . . . . . . . . . . . . . . 31 2.2.3 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . 32 2.3 Beta Estimator for 0-Tuple-Situations . . . . . . . . . . . . . . 33 2.3.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 33 2.3.2 Beta Distribution in Non-0-TS . . . . . . . . . . . . . . . . . 35 2.3.3 Parameter Estimation in 0-TS . . . . . . . . . . . . . . . . . . 37 2.3.4 Selectivity Estimation and Predicate Ordering . . . . . . . . . 39 2.3.5 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.4 Customized Sampling Techniques . . . . . . . . . . . . . . . . . . 53 2.4.1 Focused Sampling . . . . . . . . . . . . . . . . . . . . . . . . 54 2.4.2 Conditional Sampling . . . . . . . . . . . . . . . . . . . . . . 56 2.4.3 Zone Pruning . . . . . . . . . . . . . . . . . . . . . . . . . . 58 2.4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 3 JOIN STAGE: LOGICAL ENUMERATION 3.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.1.1 Point Estimates . . . . . . . . . . . . . . . . . . . . . . . . 63 3.1.2 Join Cardinality Upper Bound . . . . . . . . . . . . . . . . . . 64 3.2 Upper Bound Join Enumeration with Synopsis (UES) . . . . . . . . . 66 3.2.1 U-Block: Simple Upper Bound for Joins . . . . . . . . . . . . . 67 3.2.2 E-Block: Customized Enumeration Scheme . . . . . . . . . . . . . 68 3.2.3 UES Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 3.3.1 General Performance . . . . . . . . . . . . . . . . . . . . . . 72 3.3.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4 JOIN STAGE: PHYSICAL OPERATOR SELECTION 4.1 Operator Selection vs Join Ordering . . . . . . . . . . . . . . . 77 4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 4.2.1 Adaptive Query Processing . . . . . . . . . . . . . . . . . . . 80 4.2.2 Bandit Optimizer (Bao) . . . . . . . . . . . . . . . . . . . . . 81 4.3 TONIC: Learned Physical Join Operator Selection . . . . . . . . . 82 4.3.1 Query Execution Plan Synopsis (QEP-S) . . . . . . . . . . . . . 83 4.3.2 QEP-S Life-Cycle . . . . . . . . . . . . . . . . . . . . . . . . 84 4.3.3 QEP-S Design Considerations . . . . . . . . . . . . . . . . . . 87 4.4 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 4.4.1 Performance Factors . . . . . . . . . . . . . . . . . . . . . . 90 4.4.2 Rate of Improvement . . . . . . . . . . . . . . . . . . . . . . 92 4.4.3 Data Shift . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 4.4.4 TONIC - Runtime Traits . . . . . . . . . . . . . . . . . . . . . 97 4.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 5 TWO-STAGE OPTIMIZER FRAMEWORK 5.1 Upper-Bound-Driven Join Ordering Component . . . . . . . . . . . . 101 5.2 Physical Operator Selection Component . . . . . . . . . . . . . . 103 5.3 Example Query Optimization . . . . . . . . . . . . . . . . . . . . 103 6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 A APPENDIX A.1 Basics of Query Execution . . . . . . . . . . . . . . . . . . . . 123 A.2 Why Q? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 A.3 0-TS Proof of Unbiased Estimate . . . . . . . . . . . . . . . . . 125 A.4 UES Upper Bound Property . . . . . . . . . . . . . . . . . . . . . 127 A.5 TONIC – Selectivity-Aware Branching . . . . . . . . . . . . . . . 128 A.6 TONIC – Sequences of Query Execution . . . . . . . . . . . . . . . 129

De nos jours, des masses de données sont publiées à grande échelle dans des formats numériques. Une part importante de ces données a une structure complexe, typiquement organisée sous la forme d'arbres (les documents du web, comme HTML et XML, étant les plus représentatifs) ou de graphes (en particulier, les bases de données du Web Sémantique structurées en graphes, et exprimées en RDF). Exploiter ces données complexes, qu'elles soient dans un format d'accès Open Data ou bien propriétaire (au sein d'une compagnie), présente un grand intérêt. Le faire de façon efficace pour de grands volumes de données reste encore un défi. Les vues matérialisées sont utilisées depuis longtemps pour améliorer considérablement l'évaluation des requêtes. Le principe est q'une vue stocke des résultats pre-calculés qui peuvent être utilisés pour évaluer (une partie d') une requête. L'adoption des techniques de vues matérialisées dans le contexte de données du web que nous considérons est particulièrement exigeante à cause de la complexité structurelle et sémantique des données. Cette thèse aborde deux problèmes liés à la gestion des données du web basée sur des vues matérialisées. D'abord, nous nous concentrons sur le problème de sélection des vues pour des ensembles de requêtes RDF. Nous présentons un algorithme original qui, basé sur un ensemble de requêtes, propose les vues les plus appropriées à matérialiser dans la base des données. Ceci dans le but de minimiser à la fois les coûts d'évaluation des requêtes, de maintenance et de stockage des vues. Bien que les requêtes RDF contiennent typiquement un grand nombre de jointures, ce qui complique le processus de sélection de vues, notre algorithme passe à l'échelle de centaines de requêtes, un nombre non atteint par les méthodes existantes. En outre, nous proposons des techniques nouvelles pour tenir compte des données implicites qui peuvent être dérivées des schémas RDF sans complexifier davantage la sélection des vues. La deuxième contribution de notre travail concerne la réécriture de requêtes en utilisant des vues matérialisées XML. Nous commençons par identifier un dialecte expressif de XQuery, correspondant aux motifs d'arbres avec des jointures sur la valeur, et nous étudions des propriétés importantes de ces requêtes, y compris l'inclusion et la minimisation. En nous fondant sur ces notions, nous considérons le problème de trouver des réécritures minimales et équivalentes d'une requête exprimée dans ce dialecte, en utilisant des vues matérialisées exprimées dans le même dialecte, et nous fournissons un algorithme correct et complet à cet effet. Notre travail dépasse l'état de l'art en permettant à chaque motif d'arbre de renvoyer un ensemble d'attributs, en prenant en charge des jointures sur la valeur entre les motifs, et en considérant des réécritures qui combinent plusieurs vues. Enfin, nous montrons comment notre méthode de réécriture peut être appliquée dans un contexte distribué, pour la dissémination efficace d'un corpus de documents XML annotés en RDF. / Data is being published in digital formats at very high rates nowadays. A large share of this data has complex structure, typically organized as trees (Web documents such as HTML and XML being the most representative) or graphs (in particular, graph-structured Semantic Web databases, expressed in RDF). There is great interest in exploiting such complex data, whether in an Open Data access model or within companies owning it, and efficiently doing so for large data volumes remains challenging. Materialized views have long been used to obtain significant performance improvements when processing queries. The principle is that a view stores pre-computed results that can be used to evaluate (possibly part of) a query. Adapting materialized view techniques to the Web data setting we consider is particularly challenging due to the structural and semantic complexity of the data. This thesis tackles two problems in the broad context of materialized view-based management of Web data. First, we focus on the problem of view selection for RDF query workloads. We present a novel algorithm, which, based on a query workload, proposes the most appropriate views to be materialized in the database, in order to minimize the combined cost of query evaluation, view maintenance and view storage. Although RDF query workloads typically feature many joins, hampering the view selection process, our algorithm scales to hundreds of queries, a number unattained by existing approaches. Furthermore, we propose new techniques to account for the implicit data that can be derived by the RDF Schemas and which further complicate the view selection process. The second contribution of our work concerns query rewriting based on materialized XML views. We start by identifying an expressive dialect of XQuery, corresponding to tree patterns with value joins, and study some important properties for these queries, such as containment and minimization. Based on these notions, we consider the problem of finding minimal equivalent rewritings of a query expressed in this dialect, using materialized views expressed in the same dialect, and provide a sound and complete algorithm for that purpose. Our work extends the state of the art by allowing each pattern node to return a set of attributes, supporting value joins in the patterns, and considering rewritings which combine many views. Finally, we show how our view-based query rewriting algorithm can be applied in a distributed setting, in order to efficiently disseminate corpora of XML documents carrying RDF annotations.

Mit dem Resource Description Framework (RDF) können Eigenschaften von und die Beziehungen zwischen Ressourcen maschinenverarbeitbar beschrieben werden. Dadurch werden diese Daten für Maschinen zugänglicher und können unter anderem automatisch Daten zu einer Ressource lokalisieren und verarbeiten, unterschiedliche Bedeutungen einer Zeichenkette erkennen und implizite Informationen ableiten. Das Datenmodell von RDF und der zugehörigen Anfragesprache SPARQL basiert auf gerichteten und beschrifteten Multigraphen. Forschungsergebnisse haben gezeigt, dass relationale DBMS zum Verwalten von RDF-Daten ungeeignet sind. Native basierende RDF-DBMS können Anfragen in kürzerer Zeit verarbeiten. Der Leistungsgewinn wird durch redundantes Speichern von Tripeln in mehreren B+-Bäumen erzielt. Jedoch sind Join-ähnliche Operationen zum Berechnen des Ergebnisses erforderlich, was bei größeren Anfragen zu Leistungseinbußen führt. In dieser Arbeit wird der Resource Centered Store (RCS) entwickelt, dessen Speichermodell RDF-inhärente Eigenschaften ausnutzt, um Anfragen ohne die Notwendigkeit redundanter Speicherung effizient beantworten zu können. Die grundlegende Idee des RCS-Speichermodells besteht im Gruppieren der Daten als sternförmigen Teilgraphen auf Datenbankseiten. Die verwendeten Prinzipien ähnelt denen in RDBMS und daher können deren Algorithmen zur Beantwortung von Anfragen wiederverwendet werden. Darüber hinaus werden Transformationsregeln und Heuristiken zum Optimieren von SPARQL-Anfragen zum Finden eines möglichst optimalen Ausführungsplans definiert. In diesem Kontext wurden auch graphmusterbasierte Indexe spezifiziert und deren Nutzen für die Verarbeitung von Anfragen untersucht. Das RCS-Speichermodell wurde prototypisch implementiert und im Vergleich zum nativen RDF-DBMS Jena TDB evaluiert. Die durchgeführten Experimenten zeigen, dass das System insbesondere für das Beantworten von Anfragen mit großen sternförmigen Teilmustern geeignet ist. / The Resource Description Framework (RDF) is the conceptual foundation for representing properties of real-world or virtual resources and describing the relationships between them. Standards based on RDF allow machines to access and process information automatically and locate additional data about resources. It also supports the discovery of relationships between concepts. The smallest information unit in RDF are triples which form a directed labeled multi-graph. The query language SPARQL is also based on a graph model which makes it difficult for relational DBMS to store and query RDF data efficiently. The most performant DBMS for managing and querying RDF data implement a RDF-specific storage model based on a set of B+ tree indexes. The key disadvantages of these systems are the increased usage of secondary storage in cause of redundantly stored triples as well as the necessity of expensive join operation to compute the solutions of a SPARQL query. In this work we develop and describe the Resource Centered Store which exploits RDF inherent characteristics to avoid the requirement for storing triples redundantly while improving the query performance of larger queries. In the RCS storage model triples are grouped by their first component (subject) and storing these star-shaped subgraphs on database pages -- similar to relational DBMS. As a result the RCS can benefit from principles and algorithms that have been developed in the context of relational databases. Additionally, we defined transformation rules and heuristics to optimize SPARQL queries and generate an efficient query execution plan. In this context we also defined graph pattern based indexes and investigated their benefits for computing the solutions of queries. We implemented the RCS storage model prototypically and compared it to the native RDF DBMS Jena TDB. Our experiments showed that our storage model is especially suited to speed up the query performance of large star-shaped graph pattern.

Os operadores convencionais para comparação de dados por igualdade e por relação de ordem total não são adequados para o gerenciamento de dados complexos como, por exemplo, os dados multimí?dia (imagens, áudio, textos longos), séries temporais e seqüências genéticas. Para comparar dados desses tipos, o grau de similaridade entre suas instâncias é, em geral, o fator mais importante sendo, portanto, indicado que as operações de consulta sejam realizadas utilizando os chamados operadores por similaridade. Existem operadores de busca por similaridade tanto unários quanto binários. Os operadores unários são utilizados para implementar operações de seleção, enquanto os operadores binários destinam-se a operações de junção. A álgebra relacional, usada nos Sistemas de Gerenciamento de Bases de Dados Relacionais, não provê suporte para expressar critérios de busca por similaridade. Para suprir esse suporte, está em desenvolvimento no Grupo de Bases de Dados e Imagens (GBdI-ICMC-USP) uma extensão à álgebra relacional que permite representar as consultas por similaridade em expressões algébricas. Esta dissertação incorpora-se nesse empreendimento, abordando o tratamento aos operadores unários por similaridade na álgebra, bem como a implementação do otimizador de consultas por similaridade no SIREN (Similarity Retrieval Engine) para que as consultas por similaridade possam ser respondidas pelos Sistemas de Gerenciamento de Bases de Dados relacionais / Conventional operators for data comparison based on exact matching and total order relations are not appropriate to manage complex data, such as multimedia data (e.g. images, audio and large texts), time series and genetic sequences. In fact, the most important aspect to compare complex data is usually the similarity degree between instances, leading to the use of similarity operators to perform search and retrieval operations. Similarity operators can be classified as unary or as binary, respectively used to implement selection operations and joins. However, the Relation Algebra, employed in Relational Database Management Systems (DBMS), does not provide resources to express similarity search criteria. In order to fulfill this lack of support, an extension to the Relational Algebra is under development at GBdI-ICMC-USP (Grupo de Bases de Dados e Imagens), aiming to represent similarity queries in algebraic expressions. This work contributes to such an effort by dealing with unary similarity operators in Relational Algebra and by developing a similarity query optimizer for SIREN (Similarity Retrieval Engine), therefore allowing similarity queries to be answered by Relational DBMS

RDF et SPARQL se sont imposés comme modèle de données et langage de requêtes standard pour décrire et interroger les données sur la Toile. D’importantes quantités de données RDF sont désormais disponibles, sous forme de jeux de données ou de méta-données pour des documents semi-structurés, en particulier XML. La coexistence et l’interdépendance grandissantes entre RDF et XML rendent de plus en plus pressant le besoin de représenter et interroger ces données conjointement. Bien que de nombreux travaux couvrent la production et la publication, manuelles ou automatiques, d’annotations pour données semi-structurées, peu de recherches ont été consacrées à l’exploitation de telles données. Cette thèse pose les bases de la gestion de données hybrides XML-RDF. Nous présentons XR, un modèle de données accommodant l’aspect structurel d’XML et la sémantique de RDF. Le modèle est suffisamment général pour représenter des données indépendantes ou interconnectées, pour lesquelles chaque nœud XML est potentiellement une ressource RDF. Nous introduisons le langage XRQ, qui combine les principales caractéristiques des langages XQuery et SPARQL. Le langage permet d’interroger la structure des documents ainsi que la sémantique de leurs annotations, mais aussi de produire des données semi-structurées annotées. Nous introduisons le problème de composition de requêtes dans le langage XRQ et étudions de manière exhaustive les techniques d’évaluation de requêtes possibles. Nous avons développé la plateforme XRP, implantant les algorithmes d’évaluation de requêtes dont nous comparons les performances expérimentalement. Nous présentons une application reposant sur cette plateforme pour l’annotation automatique et manuelle de pages trouvées sur la Toile. Enfin, nous présentons une technique pour l’inférence RDFS dans les systèmes de gestion de données RDF (et par extension XR). / Since the beginning of the Semantic Web, RDF and SPARQL have become the standard data model and query language to describe resources on the Web. Large amounts of RDF data are now available either as stand-alone datasets or as metadata over semi-structured documents, typically XML. The ability to apply RDF annotations over XML data emphasizes the need to represent and query data and metadata simultaneously. While significant efforts have been invested into producing and publishing annotations manually or automatically, little attention has been devoted to exploiting such data. This thesis aims at setting database foundations for the management of hybrid XML-RDF data. We present a data model capturing the structural aspects of XML data and the semantics of RDF. Our model is general enough to describe pure XML or RDF datasets, as well as RDF-annotated XML data, where any XML node can act as a resource. We also introduce the XRQ query language that combines features of both XQuery and SPARQL. XRQ not only allows querying the structure of documents and the semantics of their annotations, but also producing annotated semi-structured data on-the-fly. We introduce the problem of query composition in XRQ, and exhaustively study query evaluation techniques for XR data to demonstrate the feasibility of this data management setting. We have developed an XR platform on top of well-known data management systems for XML and RDF. The platform features several query processing algorithms, whose performance is experimentally compared. We present an application built on top of the XR platform. The application provides manual and automatic annotation tools, and an interface to query annotated Web page and publicly available XML and RDF datasets concurrently. As a generalization of RDF and SPARQL, XR and XRQ enables RDFS-type of query answering. In this respect, we present a technique to support RDFS-entailments in RDF (and by extension XR) data management systems.

Nous explorons différents aspects des entrepôts de données et d’OLAP, le point commun de nos recherches étant le modèle fonctionnel pour l'analyse de données. Notre objectif principal est d'utiliser ce modèle dans l'étude de trois aspects différents, mais liés:- l'optimisation de requêtes par réécriture et la gestion du cache,- la visualisation du résultat d'une requête OLAP,- le mapping d'un schéma relationnel en BCNF vers un schéma fonctionnel. L'optimisation de requêtes et la gestion de cache sont des problèmes cruciaux dans l'évaluation de requêtes en général, et les entrepôts de données en particulier; et la réécriture de requêtes est une des techniques de base pour l'optimisation de requêtes. Nous établissons des conditions d'implication de requêtes analytiques, en utilisant le pré-ordre partiel sur l'ensemble de requêtes, et nous définissons un algorithme sain et complet de réécriture ainsi que une stratégie de gestion de cache optimisée, tous les deux basés sur le modèle fonctionnel.Le deuxième aspect important que nous explorons dans cette thèse est celui de la visualisation du résultat. Nous démontrons l'importance pour la visualisation de reproduire des propriétés essentielles de données qui sont les dépendances fonctionnelles. Nous montrons que la connexion, existante entre les données et leur visualisation, est précisément la connexion entre leurs représentations fonctionnelles. Nous dérivons alors un cadre technique, ayant pour objectif d'établir une telle connexion pour un ensemble de données et un ensemble de visualisations. En plus d'analyse du processus de visualisation, nous utilisons le modèle fonctionnel comme un guide pour la visualisation interactive, et définissons ce qu'on appelle la visualisation paramétrique. Le troisième aspect important de notre travail est l'expérimentation des résultats obtenus dans cette thèse. Les résultats de cette thèse peuvent être utilisés afin d’analyser les données contenues dans une table en Boyce-Codd Normal Form (BCNF), étant donné que le schéma de la table peut être transformé aisément en un schéma fonctionnel. Nous présentons une telle transformation (mapping) dans cette thèse. Une fois le schéma relationnel transformé en un schéma fonctionnel, nous pouvons profiter des résultats sur l'optimisation et la visualisation de requêtes. Nous avons utilisé cette transformation dans l’implémentation de deux prototypes dans le cadre de deux projets différents. / In this thesis, we explore different aspects of Data Warehousing and OLAP, the common point of our proposals being the functional model for data analysis. Our main objective is to use that model in studying three different, but related aspects:- query optimization through rewriting and cache management,- query result visualization,- mapping of a relational BCNF schema to a functional schema.Query optimization and cache management is a crucial issue in query processing in general, and in data warehousing in particular; and query rewriting is one of the basic techniques for query optimization. We establish derivability conditions for analytic functional queries, using a partial pre-order over the set of queries. Then we provide a sound and complete rewriting algorithm, as well as an optimized cache management strategy, both based on the underlying functional model.A second important aspect that we explore in the thesis is that of query result visualization. We show the importance for the visualization to reflect such essential features of the dataset as functional dependencies. We show that the connection existing between data and visualization is precisely the connection between their functional representations. We then define a framework, whose objective is to establish such a connection for a given dataset and a set of visualizations. In addition to the analysis of the visualization process, we use the functional data model as a guide for interactive visualization, and define what we call a parametric visualization. A third important aspect of our work is experimentation with the results obtained in the thesis. In order to be able to analyze the data contained in a Boyce-Codd Normal Form (BCNF) table, one can use the results obtained in this thesis, provided that the schema of the table can be mapped to a functional schema. We present such a mapping in this thesis. Once the relational schema has been transformed into a functional schema, we can take advantage of the query optimization and result visualization results presented in the thesis. We have used this transformation in the implementation of two prototypes in the context of two different projects.

Enterprise Database Management Systems (DBMSs) have to contend with resource-intensive and time-varying workloads, making them well-suited candidates for migration to cloud plat-forms { specifically, they can dynamically leverage the resource elasticity while retaining affordability through the pay-as-you-go rental interface. The current design of database engine components lays emphasis on maximizing computing efficiency, but to fully capitalize on the cloud's benefits, the outlays of these computations also need to be factored into the planning exercise. In this thesis, we investigate this contemporary problem in the context of industrial-strength deployments of relational database systems on real-world cloud platforms. Specifically, we consider how the traditional metric used to compare query execution plans, namely response-time, can be augmented to incorporate monetary costs in the decision process. The challenge here is that execution-time and monetary costs are adversarial metrics, with a decrease in one entailing a rise in the other. For instance, a Virtual Machine (VM) with rich physical resources (RAM, cores, etc.) decreases the query response-time, but is expensive with regard to rental rates. In a nutshell, there is a tradeoff between money and time, and our goal therefore is to identify the VM that others the best tradeoff between these two competing considerations. In our study, we pro le the behavior of money versus time for a given query, and de ne the best tradeoff as the \knee" { that is, the location on the pro le with the minimum Euclidean distance from the origin. To study the performance of industrial-strength database engines on real-world cloud infrastructure, we have deployed a commercial DBMS on Google cloud services. On this platform, we have carried out extensive experimentation with the TPC-DS decision-support benchmark, an industry-wide standard for evaluating database system performance. Our experiments demonstrate that the choice of VM for hosting the database server is a crucial decision, because: (i) variation in time and money across VMs is significant for a given query, (ii) no one VM offers the best money-time tradeoff across all queries. To efficiently identify the VM with the best tradeoff from a large suite of available configurations, we propose a technique to characterize the money-time pro le for a given query. The core of this technique is a VM pruning mechanism that exploits the property of partially ordered set of the VMs on their resources. It processes the minimal and maximal VMs of this poset for estimated query response-time. If the response-times on these extreme VMs are similar, then all the VMs sandwiched between them are pruned from further consideration. Otherwise, the already processed VMs are set aside, and the minimal and maximal VMs of the remaining unprocessed VMs are evaluated for their response-times. Finally, the knee VM is identified from the processed VMs as the one with the minimum Euclidean distance from the origin on the money-time space. We theoretically prove that this technique always identifies the knee VM; further, if it is acceptable to and a \near-optimal" knee by providing a relaxation-factor on the response-time distance from the optimal knee, then it is also capable of finding more efficiently a satisfactory knee under these relaxed conditions. We propose two favors of this approach: the first one prunes the VMs using complete plan information received from database engine API, and named as Plan-based Identification of Knee (PIK). On the other hand, to further increase the efficiency of the identification of the knee VM, we propose a sub-plan based pruning algorithm called Sub-Plan-based Identification of Knee (SPIK), which requires modifications in the query optimizer. We have evaluated PIK on a commercial system and found that it often requires processing for only 20% of the total VMs. The efficiency of the algorithm is further increased significantly, by using 10-20% relaxation in response-time. For evaluating SPIK , we prototyped it on an open-source engine { Postgresql 9.3, and also implemented it as Java wrapper program with the commercial engine. Experimentally, the processing done by SPIK is found to be only 40% of the PIK approach. Therefore, from an overall perspective, this thesis facilitates the desired migration of enterprise databases to cloud platforms, by identifying the VM(s) that offer competitive tradeoffs between money and time for the given query.

Os operadores convencionais para comparação de dados por igualdade e por relação de ordem total não são adequados para o gerenciamento de dados complexos como, por exemplo, os dados multimí?dia (imagens, áudio, textos longos), séries temporais e seqüências genéticas. Para comparar dados desses tipos, o grau de similaridade entre suas instâncias é, em geral, o fator mais importante sendo, portanto, indicado que as operações de consulta sejam realizadas utilizando os chamados operadores por similaridade. Existem operadores de busca por similaridade tanto unários quanto binários. Os operadores unários são utilizados para implementar operações de seleção, enquanto os operadores binários destinam-se a operações de junção. A álgebra relacional, usada nos Sistemas de Gerenciamento de Bases de Dados Relacionais, não provê suporte para expressar critérios de busca por similaridade. Para suprir esse suporte, está em desenvolvimento no Grupo de Bases de Dados e Imagens (GBdI-ICMC-USP) uma extensão à álgebra relacional que permite representar as consultas por similaridade em expressões algébricas. Esta dissertação incorpora-se nesse empreendimento, abordando o tratamento aos operadores unários por similaridade na álgebra, bem como a implementação do otimizador de consultas por similaridade no SIREN (Similarity Retrieval Engine) para que as consultas por similaridade possam ser respondidas pelos Sistemas de Gerenciamento de Bases de Dados relacionais / Conventional operators for data comparison based on exact matching and total order relations are not appropriate to manage complex data, such as multimedia data (e.g. images, audio and large texts), time series and genetic sequences. In fact, the most important aspect to compare complex data is usually the similarity degree between instances, leading to the use of similarity operators to perform search and retrieval operations. Similarity operators can be classified as unary or as binary, respectively used to implement selection operations and joins. However, the Relation Algebra, employed in Relational Database Management Systems (DBMS), does not provide resources to express similarity search criteria. In order to fulfill this lack of support, an extension to the Relational Algebra is under development at GBdI-ICMC-USP (Grupo de Bases de Dados e Imagens), aiming to represent similarity queries in algebraic expressions. This work contributes to such an effort by dealing with unary similarity operators in Relational Algebra and by developing a similarity query optimizer for SIREN (Similarity Retrieval Engine), therefore allowing similarity queries to be answered by Relational DBMS