Large-Scale RDF Processing with MapReduce - Large Scale and Big Data: Processing and Management - page 175

PigSPARQL MAPSIN

1000 1000

750

100

500

10

250

1 0

1000 1500 2000 2500 3000 1000 1500 2000 2500 3000

(a)

LUBM (# universities) LUBM (# universities)

10,000 3500

2800

1000

2100

100

1400

10

700

1 0

1000 1500 2000 2500 3000 1000 1500 2000 2500 3000

(b) LUBM (# universities) LUBM (# universities)

(Multiway join optimization)

PigSPARQL MAPSIN PigSPARQL MAPSIN

10,000

1000

100

10

1

1000 1500 2000 2500 3000

3000

2000

1000

0

1000 1500 2000 2500 3000

(c)

LUBM (# universities)

FIGURE 5.8 Performance comparison for LUBM Q1 (a), Q8 (b), and Q4 (c).

access of PigSPARQL. Improvements are still present but less significant, resulting
in an up to five times faster query execution.
An open issue of the evaluation remains the actual data flow between HBase
and MapReduce, as HBase is like a black box where data distribution and parti-
tioning is handled by the system automatically. Since data locality is an important