Queries generated by modern data applications and BI tools often contain repetitive yet similar, possibly correlated subqueries. We present how the Firebolt rule-based optimizer handles a specific class of such queries that contain correlated scalar aggregate subqueries over similar inputs. We do this by crafting a set of rules that work in tandem to achieve the desired end result. We discuss each of these rules in detail and explain why our technique goes beyond the current state of the art.

Ensuring the reliability and stability of Database Management Systems (DBMS) is paramount to their success. Rigorous testing and stabilization are critical prerequisites for deploying enterprise-grade DBMS solutions. This paper introduces Fishbowl, an automated testing tool developed specifically to rigorously test Google's cloud-native database products, including AlloyDB and CloudSQL. Fishbowl's architecture prioritizes extensibility, enabling the generation of diverse, enterprise-grade workloads and failure scenarios. With over two years of operational experience, Fishbowl has proven instrumental in identifying and resolving critical issues, directly contributing to the high quality of Google's database offerings. This paper details Fishbowl's architecture, components, and demonstrates its impact through a selection of real-world case studies.

Smart manufacturing is increasingly making use of visual analytics to optimize production or to identify early problem signs [Su19]. However, current solutions and approaches require professionals, especially from the data science area, to make use of it, which is for most production companies not affordable. In this paper, we describe first a best practice to sensorize plants from the wood and beverage industry to enable smart manufacturing in general. Second, we describe a new approach that aims at providing easy-to-use visual analytics functionalities that are designed to be used directly by plant workers. Plant workers usually have encompassing experience in the production and the plant, but lack of computer experience and corresponding mathematical knowledge for data analysis. Through lowering the barriers for plant workers in performing data analysis of the IoT sensors with simplified and almost automated analysis functions would give them the ability to gain insights into the production and achieve similar production optimizations and problem preventions as data science experts could. The main contributions of this article are on the one hand the best practice of how production lines of the wood and beverage industry could be made ready for smart manufacturing, but also an approaches that enable non-data scientists, especially plant workers, to perform sufficient analysis about optimal production settings and early problem cause identification.

The Ocient Hyperscale Data Warehouse is a massively parallel processing (MPP) system designed to efficiently store and analyze petabyte-scale datasets. Ocient utilizes a compute-adjacent storage architecture (CASA), where storage and compute resources are co-located to minimize data movement, thus enhancing query performance. We present the system architecture and dive deeper into data storage in segments, which do not only store columnar table data but also index information. This design, combined with parallel query processing, allows for high throughput and low-latency execution.

Beyond that, the paper highlights OcientGeo – deeply integrated data types for geospatial analytics – as well as OcientML – a machine learning integration for running analytics and model training directly inside the database system. These features expand the system’s utility across diverse industries and applications.

Modern data analytics cover a wide range of tools and frameworks besides traditional SQL-based analytics, leading to the need to transform traditional data warehouse solutions into data platforms. In this paper we focus on ecosystem integration, query performance and ease-of-use as key differentiators. We spotlight capabilities we introduced into our analytical database system Actian Vector with respect to these differentiators and highlight challenges we faced when applying research ideas to a production-ready system. The described capabilities cover ecosystem integrations with Apache Spark and a containerized services framework, cost model optimizations for robust query planning and several features to improve usability and manageability of the system.

In today's enterprises, huge amounts of data are generated every day. By applying data-driven analysis techniques, this data can be exploited for insights that may help to optimize business processes. For the systematic collection, management and processing of data, data platforms such as data warehouses and data lakes have been established, which provide a central access point of data for analytics applications. However, industrial practice shows that analytics architectures of large enterprises are not organized centrally, but instead consist of many differently shaped data platforms that are operated by separate business units for reasons such as an improved time to market, flexibility and customer orientation. Due to this heterogeneity, the sharing of data between different business units becomes difficult from both a technical and organizational perspective. Data products constitute a promising approach to address this issue, as they pursue to package data as standardized, self-contained and ready-to-consume units. This paper discusses the challenges of distributed data platform landscapes in enterprises regarding data sharing from an industry perspective and illustrates why practitioners are placing high hopes in data products to overcome these boundaries.

HUK-COBURG relies on decision trees to classify e-mails to topics.

As the data is stored within relational database systems, this work investigates on the performance gain of in-database evaluation of decision trees.

The paper first summarises the theory behind binary decision trees and describes the implementation as SQL case-when statement afterwards.

The evaluation shows the benefit of in-database application of decision trees by eliminating costful extraction-transform-load time.

In the future, we intend to optimise training as well as inference of decision trees in SQL to fully eliminate the need for data extraction.