Modeling P2P Grid Information Services with Colored Petri Nets - Grid Computing: Infrastructure, Service, and Applications - page 340

Start

Is node
leaving/
dead?

Do not accept message

Yes

No

Is node
busy?

Add message to buffer

Yes

No

Consume the required
resources, and set
to state “busy”

Populate next message
from buffer

No

Process the message

Is buffer
empty?

Return the resources

Yes

Set to “free”
state

End

FIGURE 13.6 PIndex l ow chart.

In addition, the dynamic sectioning of the PIndex algorithm was limited
to two per hop (the allowed minimum for dynamic sectioning).
First, the scalability of PIndex was evaluated. The network population
was varied with different numbers of PGs, of which each PG consists of 100
computing nodes. The behavior of PIndex under increasing loads could be
observed. Increments of 10, from 10 to 100 PGs, were simulated. However,
for clarity only simulations of 10, 20, 60, and 80 are presented.
Second, an investigation into improving the behavior of PIndex was
carried out. By grouping nodes in accordance to the activity (failure
probability), the behavior of PIndex under varying network conditions
could be observed. Not only would this investigation i nd a recom-
mended PIndex coni guration, but by grouping nodes with high failure
rates it tests the ability of PIndex to handle churn. In this simulation the
number of PGs was i xed at 100, each containing 100 nodes. Three group-
ing methodologies were investigated: (1) alike—grouping nodes with the
same failure rates; (2) even—grouping nodes with a normal distribution;