Design And Implementation R-Tree Spatial Index In MongoDB Database

With the rapid development of sensor,information technologies and the advancement of smart city,different industries and department has produced the so called Geo-Big-Data not only with big volume,but also with big velocity and variety.Effective methods to manage and retrieve such data is the key to all GIS applications features on spatial analysis,spatial data mining and decision making.With the expansion of spatial data support,RDBMS(Relational Database Manage System)is usually used for storing and indexing spatial data due to its excellent ACID features.However,as the data volume keep climbing up,traditional relation database struggles in face of scalability,throughout put and query performance.Rising with the Web2.0 application,NoSQL(No Only SQL)databases are ideal solutions to these kinds of application scenarios due to its free scalability,schema free design and high performance.MongoDB is one of the most famous NoSQL databases due to its rich query language and excellent performance.However,in spatial domain,MongoDB's native index "2d" failed to support many geometry types such as line string and polygon and "2dsphere" index only supports data with geodetic coordinate system.Querying data with 2dsphere index requires spherical computing,which is unnecessary and time consuming specially in small range.However,many applications,particularly at the city/country scale,that are still used to apply planar Cartesian coordinates to measure spatial data because of convenient acquisition and straightforward calculation.Consider the fact that R-tree is well performed in planar Cartesian coordinate system,the main purpose of this paper is to bring R-tree into this MongoDB distribute environment:1)Study MongoDB's architecture,analysis and design an efficient flattened structure to store R-tree index based on the MongoDB's document based data model.Furthermore,a database schema which is capable of manage the R-tree index system is proposed.2)Study different sharding strategies used in distribute R-tree and analysis its impact on index performance.Propose a novel sharing strategy which takes the spatial locality into account while doing load balance based on MongoDB's sharding architecture.3)Enumerate and compare different integration strategies to integrate R-tree with MongoDB seamlessly.By discussing integration details in message layer,index logic layer and R-tree layer,we implement a MongoDB's R-tree capable prototype.4)Build MongoDB cluster and load it with real world dataset.Test and compare the time consumption of operations with different indexes and sharding strategies and evaluate the systems performance.