Modeling P2P Grid Information Services with Colored Petri Nets - Grid Computing: Infrastructure, Service, and Applications - page 336

13.4.1 Existing Grid Simulators

This section gives a brief review of some existing simulators that might be
suitable for simulations of PIndex.

13.4.1.1 SimGrid

SimGrid [12] offers the freedom to design a network topology. However,
its dei nition of service times and use exists in only two static classes,
which may become problematic when modeling PIndex. PIndex has mul-
tiple PGs and vast simulated population data that will become a bottle-
neck using SimGrid. In addition, the nature of PIndex is dynamic which
cannot be modeled through the rigid dei nition of only tasks and resources;
an accommodation of P2P aspects such as dynamic updates, search algo-
rithms, and dynamic node presences must be held.

13.4.1.2 GridSim

GridSim [13] builds on top of SimJava [14], which is primarily an API to
Java threads with a built-in discrete clock. However, the constraints placed
in accessing these threads pose some l exibility issues when trying to
implement prioritizations for different job types and to mimic the parallel
nature of the PGs of PIndex. In addition, similar to SimGrid, having a cen-
tral controller class may reduce performance when simulating many
thousands of nodes as required by PIndex.

13.4.1.3 GangSim

Although created specii cally for Ganglia [15], GangSim [16] does high-
light the need for using real data to improve the plausibility of simulated
results. However, it is unsuitable for simulating PIndex as its precision is
derived on the assumption that jobs run for a minimum of 100 seconds,
which is too long a period for an information service.

13.4.1.4 CPN Tools Package

Based on the theory of colored Petri nets (CPNs), the CPN Tools package
[17] enables one to create a mathematical model from a bipartite graph.
Howe ve r, PI nde x's mu lt i-PG n at u r e c a n not b e si mu l at e d w it h t h i s pac k age,
as the package was designed for standard l at Petri nets, where a single
model represented the whole network dictating that PIndex should have a
static architecture, which is not the case.
Summarizing, existing simulators usually work under the assumption
that a grid structure is l at and lacks dynamism (i.e., i xed and predei ned).