Algebraic Optimization of RDF Graph Pattern Queries on MapReduce - Large Scale and Big Data: Processing and Management - page 208

6.7 CASE STUDY: EVALUATION OF NTGA EXECUTION PLANS
In this section, we present results comparing the performance of evaluating graph
pattern queries using NTGA execution plans against relational-style execution plans
in Apache Pig. We evaluated scalability in terms of number of join operations and
the size of the cluster.

6.7.1 s etuP anD t estbeD
Experiments were conducted on 5- to 30-node Hadoop clusters. Two synthetic data
sets were used—data set D 1 with size 51 GB (approximately 200 million n-triples)
generated using the BSBM benchmark generator and D 2—modiied version of the
analysis benchmark data set used in [34] with size 43 GB (approximately 1 billion
3-a r y tr iples).

6.7.2 s Calability with i nCreasing J oins
Figure 6.11a shows the results of queries—Q1, Q2, Q3, Q4 with 3, 5, 9, and 11 joins,
respectively. The number of star subpatterns varies from one in Q1 to four in Q4,

(a)

7000

Pig
RAPID+

6000

5000

4000

3000

2000

1000

0

Q1 Q2 Q3 Q4

(b)

9000

Pig
RAPID+

8000

7000

6000

5000

4000

3000

2000

1000

0

10-nodes 20-nodes 30-nodes

FIGURE 6.11 Scalability study with (a) increasing joins using the 51 GB data set on a
10-node cluster and (b) increasing cluster size using the 43 GB data set.