Big Data Computing

189Linked Data in Enterprise Integration

having a radius between 49 and 99 m. Thus, running the algorithms with

a threshold of 99 m led to only a small a priori overhead and HYPPO per-

forming remarkably well. Still, even on such data distributions,

HR

3

was

able to discard even more data and to reduce the number of unnecessary

computations by more than 50% relative. In the best case (experiment 4,

α

=

32, see Figure 5.6d),

HR

3

required approximately 4.13 × 10

6

less compari-

sons than HYPPO for

α

= 32. Even for the smallest setting (experiment 1, see

Figure5.6a),

HR

3

still required 0.64 × 10

6

less comparisons.

We also measured the runtimes of SILK, HYPPO, and

HR

3

. The best run-

times of the three algorithms for each of the tasks i