Modeling P2P Grid Information Services with Colored Petri Nets - Grid Computing: Infrastructure, Service, and Applications - page 342

140

120

100

80

60

40

20

0

0 100 200 300 400 500 600 700 800

Simulated time (minutes)

10 PG 20 PG 60 PG 80 PG

FIGURE 13.8 Message usage in PIndex.

Furthermore, on observing the results no sudden “bursts” in increased
message consumption occur that can be associated with a network com-
pensating for a failed node (churn) or the creation of a network bottleneck.
Note that the leveling off of message usage after its initial startup rel ects
a state of stability.
Figure 13.9 shows the number of nodes being used as the network popu-
lation is increased. Again a linear rise of nodes being used is seen as
network population increase, proving an expected cost when implement-
ing PIndex and that the creation of “key routing node” is not formed.
Although a general increase is seen with node population, it is worth not-
ing that its gradient is not 1. Considering the case where peak node usage
for 10 PGs is 52, and for 80 PGs is 326, this gives a gradient of 6.24/8 ~ 0.78.
This means we get six times the node usage for eight times the popula-
tion, and in correlation with messages being sent no sudden “bursts” in
increased node usage can be observed. This result is most likely due to
how the PIndex search algorithm works, in that the more population that
exists, the more parallel message can be sent. This increases the overall
messages being sent but lessens the time spent on these searches, and
hence the time computing nodes spend on processing a search.
Figure 13.10 shows the plot of number of messages sent given the
different grouping methods implemented, as having an even distribution