Front Matter

Magdalena Balazinska, Xiaofang ZhouPVLDB 2019

Proceedings of the VLDB Endowment Logo

Overview

This document serves as the editorial introduction and table of contents for Volume 13, Issue 4 of the Proceedings of the VLDB Endowment (PVLDB), published in December 2019. Under the leadership of editors Magdalena Balazinska and Xiaofang Zhou, this issue presents 12 peer-reviewed articles that were featured at the 46th International Conference on Very Large Data Bases (VLDB 2020). The collection underscores a significant industry shift toward hardware-aware indexing and the integration of machine learning into core query processing.

Core Research Themes

The 12 papers in this issue address several specific bottlenecks in modern data management, focusing on the "velocity, volume, and variety" of contemporary data environments.

Hardware Evolution and Optimization

Traditional database engines often lag behind hardware advancements. This issue features four papers focused on database indexing and optimization for new hardware. A notable contribution is "DPTree," a differential indexing structure designed for the unique read/write characteristics of Persistent Memory (PM). Additionally, research explores using machine learning to learn the "hidden parameters" of Solid State Drives (SSDs) to better optimize database performance.

Neural-Database Hybrids for Video Analytics

As video content becomes a dominant data type, querying it remains computationally expensive. Two papers, BlazeIt and Panorama, propose using specialized neural networks to make video querying more declarative and efficient. This shift treats video analytics as a first-class query type, integrating neural networks directly into the database optimizer's toolkit.

Algorithmic Responsibility and Knowledge Integrity

The field is maturing to include the social dimensions of data management. One paper addresses "individual fairness" in data representations, treating social impact with the same algorithmic rigor as query latency. Another explores the integrity of knowledge bases by automatically mining "anti-knowledge" (factual mistakes) from massive, evolving sources like Wikipedia.

Methodological Diversity

The issue showcases a diverse array of methodologies, from theoretical refinements to system-level innovations:

  • Join Algorithms: New research on stream-based and sample-based join algorithms, including theoretical guides for "joins on samples" to help practitioners navigate the trade-offs in approximate query processing.
  • Graph Databases: Two papers address challenges in graph data, specifically "Hop-constrained s-t Simple Path" enumeration and "Canonical Hub Labeling" algorithms.

The PVLDB Review Model

PVLDB continues to pioneer a hybrid journal-style monthly submission cycle. This model includes a three-month revision process, ensuring that the research presented at conferences like VLDB 2020 represents deeply vetted, high-quality work. The 12 papers in this issue were selected by a board of 186 expert researchers.

Conclusion and Future Directions

While these papers provide valuable insights, they also highlight emerging fields with significant room for growth. The editors note that the integration of neural networks into query processing, for instance, remains a frontier where researchers must continue to optimize accuracy-speed trade-offs and generalizability across different video domains.

Summary updated

BibTeX
@article{balazinska2019front,
  author = {Magdalena Balazinska and Xiaofang Zhou},
  title = {{Front Matter}},
  journal = {Proc. VLDB Endow.},
  year = {2019},
}
Figure