Buffer pool translation—resolving an on-disk page identifier to an in-memory frame—was once a heavyweight operation whose bookkeeping consumed a substantial fraction of CPU cycles in early OLTP engines. Recent systems reduce this cost with techniques such as pointer swizzling, in-page hints, OS page-table mappings, prediction, sharding, and direct arrays. This paper surveys buffer pool translation mechanisms and then studies two orthogonal ways to reduce the cost of hash-table-based translation: placing related pages near each other in the frame array so neighboring pages prefer neighboring frame slots, and choosing a lookup policy that bypasses the hash table when a predicted frame validates. The benefit of each depends on workload locality, displacement, and payload size: on an in-memory translation microbenchmark, combining relation-local placement with bypass yields up to 1.46 × higher translation throughput, while the advantage shrinks to near zero once large payload reads dominate per-access cost.
Durch die enormen Datenmengen, die moderne Computersysteme effizient verarbeiten müssen, rückt der Hauptspeicher immer weiter in das Zentrum der Betrachtung bei der Softwareentwicklung, bei Herstellern von Hardware und in der Wissenschaft. Neue Technologien wie HBM („High-Bandwidth Memory“), PIM/PNM („Processing In/Near Memory“) und persistente Hauptspeichermodule werden entwickelt und in komplexen NUMA-Topologien, zum Beispiel auf Basis von CXL, verknüpft. All dies hat Auswirkungen auf den kompletten Systemsoftwarestapel und bis in die Anwendungen hinein. In diesem Sinne sind viele neuartige Hauptspeichertechnologien disruptiv. In dieser Lage ist das Ziel des im Jahr 2022 von der DFG eingerichteten Schwerpunktprogramms 2377 die objektive Bewertung dieser neuen Technologien und die Erforschung der Frage, wie sich diese auf bestehende und gänzlich neue Hardware‑/Software-Architekturen auswirken. Dabei entwickeln die Teilprojekte, die verschiedenen Forschungs-Communitys der (Kern‑)Informatik zuzurechnen sind, aktiv neuartige Lösungen, um die Potenziale der Technologien aufzuzeigen. Dieser Artikel erläutert die Hintergründe und Ziele des Programms, fasst die Ergebnisse der dreijährigen ersten Förderperiode zusammen und gibt einen Ausblick auf die zweite Förderperiode.
Traditionally, deploying database systems required acquiring and managing large, expensive servers on-premise. These servers feature a memory and storage hierarchy, including CPU caches, DRAM memory, HDDs and/or SSDs, and sometimes tape. For almost 40 years, Gray and Putzolu’s five-minute rule [4, 10, 12, 13] has helped guide system architects to the break-even point between memory caching and direct local storage access for these systems. The rise of cloud computing has significantly changed the way systems are deployed. Cloud services offer on-demand computing and storage, removing the need for companies to acquire and manage their own hardware. Furthermore, the cloud has introduced a new layer of the storage hierarchy for database systems: object stores. This durability and low cost have made object stores ubiquitous in modern cloud-native databases and data infrastructures [5, 6, 16]. We believe a rule of thumb similar to the five-minute rule is needed for object caches and storage for disaggregated cloud data system designs. However, it is not straightforward to adapt the established rules to the cloud as they presume fixed hardware, while, in the cloud, resources are dynamic and costs are determined by usage. One can argue that accessing object stores over the network incurs higher latency and lower bandwidth than directly attached storage devices, placing them at the bottom of the storage hierarchy. However, advances in cloud storage and networking have significantly impacted cost-effectiveness, latency variability, and dynamic workload optimization, challenging these traditional assumptions. For example, in non-storage-optimized AWS instances, network bandwidth often exceeds local storage read bandwidth, and writing to local storage devices tends to be half as fast as reading, while writing to S3 can match network bandwidth. This paper is a summary of our earlier work work [9] that proposes a cost model and new rules of thumb to help system designers determine when caches become cost-effective for analytical workloads in the cloud. While perhaps unsurprising, our analysis on AWS shows that caches are beneficial when a system makes (1) two requests per hour for latency-sensitive workloads, or (2) seven requests per second for non-latency-sensitive workloads. These results are consistent with and help explain the near ubiquity of object store caches in cloud analytics systems.
The smart city paradigm has evolved into a central framework for understanding digital transformation in urban contexts, yet it remains conceptually fragmented and architecturally heterogeneous. This paper addresses this dual fragmentation by systematically synthesizing 69 academic smart city definitions and contrasting them with empirical architectural patterns observed in leading European and North American cities. First, we derive an integrative definition that conceptualizes the smart city as a learning, cyber-socio-technical urban system integrating digital infrastructures, data resources, governance structures, and human capital to enhance quality of life, competitiveness, and ecological resilience. Second, we extend this conceptualization through a data- and database-oriented perspective that frames smart cities as heterogeneous urban data ecosystems governing the full data lifecycle via interoperable architectures and context-sensitive governance mechanisms. By linking normative theory with observable data-engineering practices, the study establishes a conceptual bridge between smart city discourse and urban data architecture, providing a structured foundation for future comparative research on adaptive urban data ecosystems.
Currently, the use of GenAI is the dominant topic in computer science and society. Consequently, research is momentarily underway in many areas of database technology to explore the possibilities offered by LLMs and how they can complement existing systems. However, the lack of reliability and reproducibility is a well-known problem. So what could be more natural than to discuss this topic at a conference? The Spring Symposium of the GI Specialist Group Database Systems placed this question at the center of its agenda and took place under the title Can We Trust the Model? Reproducibility and Reliability of LLM Results. The two-day symposium brought together researchers from academia and industry to discuss current developments at the intersection of database systems and artificial intelligence. The scientific program featured thirteen invited talks covering topics such as LLM-assisted database technologies, knowledge graphs, information retrieval, data quality, and database education in the era of generative AI. Complementing the main program, a poster session provided young researchers with the opportunity to present and discuss their work with the community. The symposium was further enriched by updates from the GI Specialist Group Database Systems, an introduction to the recently founded Faculty for Informatics and Data Science at the University of Regensburg, and a conference dinner sponsored by Snowflake that fostered networking and scientific exchange beyond the formal program.
Accurate short-term parking availability forecasting is an important component of intelligent transportation systems. While numerous machine learning and deep learning models have been proposed, most approaches rely on a single predictor and implicitly assume uniform performance across forecast horizons, despite evidence that accuracy varies with lead time. This paper presents a systematic evaluation of multi-step (1–24 h) parking availability forecasting using sensor-based occupancy data from two urban parking facilities. We show that different model classes exhibit complementary strengths at different horizons, with no single model consistently achieving the lowest error across the full forecast range. To address this limitation, we propose a horizon-aware hybrid forecasting approach that combines two complementary base models using validation-driven weighted blending. Two variants are considered: a Global Weighted Blend with a shared weight and a Local Weighted Blend with horizon-specific weights. Experimental results show that both hybrid approaches outperform individual baseline models, with the Local Weighted Blend achieving the best overall performance across all evaluation metrics. These findings demonstrate that horizon-aware blending provides a simple and effective strategy for improving multi-step parking availability forecasts. As an outlook, we propose a context-aware, adaptive model-management architecture for sensor-based smart city systems.
Memory tiering provides a promising approach to enhance the flexibility of system memory configurations in modern compute servers. Intel® Flat Memory Mode (FLAT) is a hardware-managed memory tiering system for CXL memory that operates at 64 B cache-line granularity. In this paper, we evaluate the performance of FLAT using On-Line Analytical Processing (OLAP) workloads in SAP HANA, an in-memory database management system. Our results demonstrate that FLAT outperforms software-managed memory tiering and directly-attached CXL memory by leveraging cache-line granularity of data placement and exploiting data locality. For the first time, we evaluate FLAT with large-scale enterprise business warehouse workloads in SAP HANA, showing that it achieves performance comparable to a DRAM-only baseline, despite the high latency and limited bandwidth of CXL memory. Currently, FLAT accepts only a 1:1 ratio between local and CXL memory. When the local memory capacity exceeds that of CXL memory, Intel® Flat Memory Mode operates in a mixed mode (MIXED) containing flat memory and additional dedicated memory with remaining local memory in a separate NUMA node. Our evaluation of MIXED shows that performance degradation depends on access patterns and the degree of data locality exploited in flat memory. However, it requires software adjustments to utilize both dedicated memory and flat memory. To eliminate the need for such adjustments in MIXED, we propose a consolidated mode (CONSOLIDATED) that merges them into a single NUMA node. Our evaluation demonstrates that CONSOLIDATED results in negligible performance degradation, regardless of the ratio between dedicated memory and flat memory, and requires no software modifications at the application level.
The proliferation and sharing of data in smart cities presents both unprecedented opportunities for urban innovation and substantial challenges around privacy, governance, and data ownership. To leverage the opportunities and handle the challenges, data trusts present a model for secure, privacy-preserving data sharing in urban environments. We critically review regulatory frameworks such as the European Union’s GDPR and Data Governance Act and perform a structured literature review and analysis of existing data governance and privacy-enhancing technologies that integrate usage control, advanced anonymization, and consent management to enable scalable, redacted data sharing while upholding individuals’ rights and intellectual property boundaries. Our results shed light on an emerging concept for resilient smart city data ecosystems and outline future research avenues for trustworthy urban data sharing and governance.
Large Language Models (LLMs) are widely used today in software development, including generating code and SQL. Agents, combined with the Model Context Protocol (MCP), enable LLMs to interact directly with database management systems and automate tasks such as schema creation, migrations, and query execution. However, effective interaction with an external system requires knowledge of specific details, such as the protocol, endpoint, and the SQL dialect to be used. The Model Context Protocol defines a standard that allows exposing tools, resources, and prompts in a discoverable, machine-readable format, enabling agents to discover the capabilities of systems and interact with them. This article explores the use of LLMs, agents, and the MCP in today’s database development. Additionally, it discusses how to implement your own MCP server, focusing on a simple MCP server for the PostgreSQL database management system that manages databases and executes SQL. Furthermore, it illustrates how an agent can utilize this MCP server through prompts.
In smart city data engineering, accurate traffic forecasting is crucial for effective urban planning and management, as well as for developing Urban Digital Twins (UDTs). Extending prior work [1] on multistep, multi-scale traffic flow forecasting, this paper explores integrating weather data into traffic forecasting models using graph neural networks (GNNs) to assess its impact. The study focuses on selecting features and time ranges for multivariate multistep traffic forecasting. It experiments with GNN setups and compares single-node per Traffic Eye Universal (TEU) configurations with combined nodes. The impact of incorporating weather features on model accuracy and training efficiency is evaluated. In the present setting, with motorized traffic in the city of Osnabrück and weather data available only at the city level, the key findings are that weather data did not improve forecasting accuracy and increased training time. Potential benefits for bicycles and pedestrians are noted as a hypothesis, as data limitations preclude further exploration. The study underscores the efficacy of correlation matrices and mutual information for feature selection, highlighting the importance of capturing non-linear relationships. It suggests that, in this configuration, the exclusive use of TEU data yields the best performance, with the node structure having a negligible impact on results. These findings imply that city-level weather data may not be a predictive value for all traffic types and motivate future research on sensor-level weather and road-surface data, as well as richer feature selection. This study contributes to the understanding of smart city data systems, underscoring the importance of efficient feature selection and the challenges of integrating diverse data sources.
Aimed at combating misinformation through Artificial Intelligence-assisted fact-checking, this paper introduces the Philippine Online Misinformation Inference (POMI) dataset, a compilation of 10,132 factual and false claims from Philippine fact-check articles, and the first misinformation corpus in the Philippines utilizing Natural Language Inference. In the benchmarking, LLaMA 3.1 70B achieved a higher accuracy of 85.1
In this short experience report, we present our attempt to integrate output from an ongoing research project with a traditional, mostly literature-based, course on media manipulation, bias and fake news, to turn an existing undergraduate course into a more engaging experience for attending students. To this end, we successfully utilized our system BiasScanner for news bias detection and classification (BiasScanner.org).
Conspiracy theories (CTs) pose significant societal challenges due to their rapid online spread and harmful real-world consequences. Most computational research relies on binary, text-based classification, which helps estimate the prevalence of conspiracist content but reveals little about how CTs are communicated and largely ignores the multimodal nature of online discourse. This study addresses these gaps with a conceptually grounded, scalable multimodal pipeline for fine-grained CT analysis. Using a corpus of German-language conspiracist Telegram messages from November 2024, the pipeline applies zero-shot LLM classification in two stages: binary CT detection and multi-label classification of text- and image-based discursive strategies, such as acts of disclosure, presenting evidence, and discrediting mainstream accounts. We further process the model outputs using lightweight lexical methods to examine strategy frequencies, characteristic terms, and common term co-occurrences, and we qualitatively analyze selected posts to illustrate how discursive strategies manifest at the message level. Our approach achieves solid binary CT classification (F1 = 0.76) and promising fine-grained strategy performance (macro-F1 = 0.51–0.64). Methodologically, we show how LLM-generated labels can be systematically processed for corpus analysis—an aspect often overlooked in prior work. Empirically, we provide insights into German-language CT discourse on Telegram during a politically turbulent period and highlight the role of visual formats such as news-style layouts, collages, and symbolic imagery. To the best of our knowledge, this is the first computational multimodal analysis of CT-related discursive strategies in German-language Telegram channels.
Article reliability measures the truthfulness of news content, providing a direct assessment of factual accuracy and credibility. Organisations such as Ad Fontes Media manually assess article reliability as one of the core components for media bias evaluation. Despite its importance, automatic article reliability classification remains underexplored, partly due to the lack of datasets with suitable annotations. In this paper, we first release a reconstructed version of an existing dataset, consisting of 5270 articles with a new four-class annotation based on article reliability. Second, we propose HT-MAGPIE, a hierarchical transformer leveraging MAGPIE—a large-scale model pre-trained on bias-related tasks—to produce bias-aware representations. We demonstrate that HT-MAGPIE outperforms fine-tuned BERT by 5.02
Subjective text classification tasks, such as abuse detection and stance analysis, often suffer from high levels of annotator disagreement. Conventional approaches typically collapse these disagreements into a single ground truth, thereby discarding valuable supervision signals. We propose MO-WEL (Multi-Objective Weighted Ensemble Learning), a novel framework that explicitly leverages annotator disagreement by jointly optimising ensemble weights and size under multiple objectives. Candidate predictors are trained on diverse label projections obtained through random sampling or annotator-specific selection, and ensemble weights are optimised with respect to three complementary losses: F1 score, cross-entropy and Manhattan distance, alongside a regularisation term. Experiments on four benchmark datasets (ConvAbuse, HS-Brexit, MD-Agreement and ArMIS) show that MO-WEL consistently outperforms strong baselines in accuracy, calibration, and distributional alignment. A case study further demonstrates that MO-WEL produces predictions that balance majority correctness with minority annotator perspectives, yielding interpretable and reliable outputs. Our findings highlight the importance of modelling annotator diversity and suggest ensemble optimisation as a principled means of incorporating disagreement into subjective NLP tasks.
Our contribution is concerned with the increasing amount of conspiracy theories and other forms of mis- or disinformation spreading on social media. We address this challenge to democratic opinion formation with a two-pronged approach. On the one hand, we identify specific conspiracy narratives in a social media corpus; on the other hand, we look at general linguistic dimensions that contribute to the overall drivel-like quality of such texts regardless of the narratives involved. For the present contribution, six distinct dimensions, along with an overall measure of drivel-like quality, were assessed using a five-point Likert scale. A sample of approximately 2000 texts drawn from German Telegram was manually annotated. We present the calculation of inter-annotator agreement to evaluate annotation consistency, conduct correlation analyses to examine the relationships between the individual dimensions, and fit a linear model to predict the overall drivel-like quality of the texts based on the individual dimensions. In addition, we train ordinal regression models to predict the values of each dimension from bag-of-n-grams representations. Finally, an analysis of feature weights identifies which n-grams serve as the most reliable indicators of each dimension.
Linguistic bias in online news and social media is widespread but difficult to measure. Yet, its identification and quantification remain difficult due to subjectivity, context dependence, and the scarcity of high-quality gold-label datasets. We aim to reduce annotation effort by leveraging pairwise comparison for bias annotation. To overcome the costliness of the approach, we evaluate more efficient implementations of pairwise comparison-based rating. We achieve this by investigating the effects of various rating techniques and the parameters of three cost-aware alternatives in a simulation environment. Since the approach can in principle be applied to both human and large language model annotation, our work provides a basis for creating high-quality benchmark datasets and for quantifying biases and other subjective linguistic aspects. The controlled simulations include latent severity distributions, distance-calibrated noise, and synthetic annotator bias to probe robustness and cost-quality trade-offs. In applying the approach to human-labeled bias benchmark datasets, we then evaluate the most promising setups and compare them to direct assessment by large language models and unmodified pairwise comparison labels as baselines. Our findings support the use of pairwise comparison as a practical foundation for quantifying subjective linguistic aspects, enabling reproducible bias analysis. We contribute an optimization of comparison and matchmaking components, an end-to-end evaluation including simulation and real-data application, and an implementation blueprint for cost-aware large-scale annotation
