HGrid: A Data Model for Large Geospatial Data Sets in HBase

Dan Han, Eleni Stroulia

2013 · 47 citations · 9 references

Concepts

TL;DR

Cloud‑based infrastructures enable massive data collection and analysis, yet NoSQL databases lack robust data organization, making systematic data‑schema design essential for scalable applications. This study targets geospatial big‑data systems and proposes the HGrid data model for HBase to address their scalability needs. HGrid combines a quad‑tree primary index with a regular grid secondary index in HBase, and its performance was compared to pure quad‑tree and grid models under uniform and skewed data. HGrid achieves scalable, efficient range and k‑nearest‑neighbor queries, offers greater flexibility for discontinuous and skewed spaces, consumes less index space than its competitors, and yields guidelines for organizing geospatial data in HBase.

Abstract

Cloud-based infrastructures enable applications to collect and analyze massive amounts of data. Whether these applications are newly developed or they are being evolved from existing RDBMS-based implementations, NoSQL databases offer an attractive platform with which to address this challenge. However, developers find it difficult to effectively manage data in NoSQL databases, because these platforms do not offer much support for data organization. Since poor data organization may abuse the features of the NoSQL database and result in unsatisfactory performance, developing a systematic method for NoSQL database data-schema design is a timely and important problem. In this paper, we focus on geospatial applications, as a family of big-data systems with distinct data types and usage patterns, in need of scalability. We propose the HGrid data model for HBase, based on a hybrid index structure, combining a quad-tree and a regular grid as primary and secondary indices correspondingly. We have comparatively evaluated the performance of HGrid with uniform and skewed data, against two other data models based on quad-tree and regular-grid indices. Our results demonstrate that HGrid scales well and supports efficient performance for range and k-nearest neighbor queries. Although this model does not outperform all its competitors in terms of query response time, it is more flexible for discontinuous and skewed space, and its index requires less space than the corresponding quad-tree and regular-grid indices, which makes its deployment possible with less resources. Through this study, we also formulate a set of guidelines on how to organize data for geospatial applications in HBase.

References

9