Similarity Join for Big Geographic Data - Geographical Information Systems: Trends and Technologies - page 45

SynthData, SF1

Output Size
MRSimJoin
MRThetaJoin 140 15000000

120

12500000

100

10000000

80

7500000

60

5000000

40

2500000

20

0 0

1 2 3 4 5

Epsilon (miles)

Fig. 10. Increasing Epsilon-SynthData.

Performance Evaluation with GeoNames

Increasing Scale Factor . Figure 11 compares the way MRSimJoin and
MRThetaJoin scale when the data size increases (SF1-SF10). The results for
GeoNames are similar to the ones we found for the case of synthetic data.
Specifi cally, the execution time of MRThetaJoin grows from being 1.4 times
of that of MRSimJoin for SF1 to 8.6 times for SF10.

Increasing Epsilon . Figure 12 shows how the execution times of MRSimJoin
and MRThetaJoin increase when epsilon increases. As in the case of
SynthData, the performance of MRSimJoin is better than the one of
MRThetaJoin for all evaluated values of epsilon. The execution time of
MRSimJoin is 64.2% of the one of MRThetaJoin for epsilon = 1 mile, and
89.8% for epsilon = 5 miles.

GeoNames, Eps:2

Output Size
MRSimJoin
MRThetaJoin

5000 250000000

4000 200000000

3000 150000000

2000 100000000

1000 50000000

0 0

123456789 0

Scale Factor (SF)

Fig. 11. Increasing SF-GeoNames.