ZNS - Efficient query processing with ZurichNoSQL

Abstract

NoSQL data stores have recently gained popularity as an alternative to relational database management systems since they typically do not require a fixed schema and scale well for large data sets. These systems have often been tuned to a number of very specific operations such as writing or reading of large data sets. However, none of these novel systems has been demonstrated to efficiently perform multi-dimensional range queries incorporating many boolean operators, a task which is commonly used in scientific data exploration, data warehousing and business analytics.In this paper we introduce ZurichNoSQL (ZNS) - a novel NoSQL main memory store that supports efficient processing of multi-dimensional point queries and range queries. The key idea of ZNS is to store the data in a column format (compressed column storage) similar to systems used in high performance computing. Moreover, the ZNS architecture is based on a set of low-level main memory techniques ensuring that CPU caches are being used efficiently. Our experimental results comparing to popular NoSQL stores such as FastBit, MongoDB and Spark SQL demonstrate that ZNS significantly outperforms these systems in most cases.