Towards Off-the-Shelf Real-Time Transactional Analytics on Cloud-Native Database Systems
Guardado en:
| Publicado en: | ProQuest Dissertations and Theses (2025) |
|---|---|
| Autor principal: | |
| Publicado: |
ProQuest Dissertations & Theses
|
| Materias: | |
| Acceso en línea: | Citation/Abstract Full Text - PDF |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Resumen: | Hybrid Transactional and Analytical Processing (HTAP) systems aim to unify transactional and analytical workloads within a single platform, enabling real-time insights over fresh data. Current HTAP solutions face two key limitations: the lack of a systematic methodology to evaluate real-time analytics capabilities, and the absence of a non-intrusive architecture that allows organizations to enable real-time analytics using their existing Transaction Processing (TP) and Analytical Processing (AP) engines without costly migrations.This dissertation addresses these challenges through two main contributions. First, we introduce HATtrick, an intuitive and systematic benchmark designed to evaluate HTAP systems across two orthogonal dimensions: throughput frontier, which captures absolute performance and the system’s ability to handle concurrent transactional and analytical workloads without interference, and freshness, which measures how up-to-date analytical query results are with respect to the most recent transactions. We also propose a visualization method that makes these metrics easy to interpret, helping users understand trade-offs and draw meaningful conclusions across systems. Our evaluation demonstrates that while modern HTAP systems have improved, substantial opportunities for optimization remain.Second, we propose HERMES, a novel off-the-shelf HTAP architecture that enables real-time transactional analytics using an organization’s existing TP and AP engines—without requiring engine modifications or expensive migrations to a new HTAP system. HERMES introduces a lightweight middle layer between the engines and storage, which dynamically merges live transaction logs with analytical reads to ensure query freshness. The design also preserves performance isolation, supports end-to-end transactional consistency, and enables fine-grained control over isolation levels for transactional analytics. We implemented a prototype using MySQL and DuckDB in the cloud and show that HERMES achieves up to 3× higher throughput on transactional analytics workloads compared to native HTAP systems.Together, these contributions provide both rigorous tools for evaluating HTAP systems and a practical architecture for enabling real-time analytics in production environments. We hope this work encourages the HTAP community to refine benchmarks, build plug-and-play solutions, and define clear design principles to make real-time analytics accessible to a broader range of organizations. |
|---|---|
| ISBN: | 9798314879894 |
| Fuente: | ProQuest Dissertations & Theses Global |