Similarity Join for Big Geographic Data - Geographical Information Systems: Trends and Technologies - page 47

GeoNames, Eps:2

1200

MRSimJoin
MRThetaJoin
Ideal

1000

800

600

400

200

0

(SF1,2) (SF2,4) (SF3,6) (SF4,8) (SF5,10)

(Scale Factor, Number of Nodes)

Fig. 13. Increasing Number of Nodes and SF-GeoNames.

GeoNames, SF5, Eps:2

600 10

NumRounds
Execution Time
Polynomial Trendline (Exec. Time)

500

8

400

6

300

4

200

2

100

0 0

15 60 105 150 195 240 285

Number of Pivots

Fig. 14. Increasing Number of Pivots-GeoNames.

Conclusions and Future Work

MapReduce-based systems have become a crucial component to effi ciently
process and analyze the large amounts of geographical data currently
available in many commercial and scientifi c organizations. The Similarity
Join is recognized as one of the most useful data analysis operations and has
been used in many application scenarios. While multiple implementation
techniques have been proposed for the Similarity Join, very little work has
addressed the study of MapReduce-based Similarity Joins for geographical
data. This chapter focuses on the study, design, and implementation
techniques of MRSimJoin, a MapReduce-based Similarity Join algorithm
that can be used with geographical data and distance functions. MRSimJoin
iteratively partitions the data until the partitions are small enough to be
effi ciently processed in a single node. Each iteration executes a MapReduce
job that processes the generated partitions in parallel. We implemented