A Data Mining Based Publish/Subscribe System over Structured Peer-to-Peer Networks - Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing - page 10

4.2 Experimental Results

4.2.1 Overhead for Matching

We first compare the numbers of stored subscriptions on RPs. In DMPSS and Ferry,
the numbers of RPs are both seven. Because the number of RPs in Eferry is large
(more than one hundred), we rank the RPs by the number of stored subscriptions,
and select the first 10 RPs for comparison. As shown in Fig. 1 and Table 1, DMPSS
realizes better even storage distribution than Ferry-Pred and Eferry, and Ferry-Rnd
has the best storage load performance. Besides, from Fig. 1 and Table 1 we can see
that although Eferry increases the number of RPs, most of subscriptions are stored
on a small part of nodes. As a result, the rest of RPs, which consume a large number
of messages for matching, only store a small amount of subscriptions. However, in
DMPSS, subscriptions that don't contain frequent itemsets are stored on only three
RPs, and the event publication cost can be reduced dramatically.

4000

DMPSS
Ferry−Pred
Ferry−Rnd
Eferry

3500

3000

2500

2000

1500

1000

500

0

1 2 3 4 5 6 7 8 9 10

RP node ranking (by the number of stored subscriptions)

Fig. 1 The number of stored subscriptions on each RP

Ta b l e 1 The number of stored subscriptions on each RP
RP 1

RP 2 RP 3 RP 4 RP 5 RP 6 RP 7 RP 8 RP 9 RP 10

DMPSS 3067 2655 2152 1320 376 339 331

Ferry-Pred 3650 3480 1610 1050 390 50 10

Ferry-Rnd 1523 1508 1456 1443 1442 1438 1430

Eferry 3096 2539 2141 1057 511 23 21 20 19 17

DMPSS+LB 3148 2614 1295 1069 1019 374 372 349