PlanAgents
Monsters
MarkovKillers
1000
500
200
50
500
1000
PlanAgents
Monsters
MarkovKillers
1000
500
200
50
500
1000
CPUs =
10
CPUs =
17
PlanAgents
Monsters
MarkovKillers
1000
500
200
50
500
1000
PlanAgents
Monsters
MarkovKillers
1000
500
200
50
500
1000
CPUs =
10
CPUs =
17
VisionRadius =
20
VisionRadius =
10
Fig. 1. 2
5
Factorial Combinations
Wall clock time response is measured using the unix shell method time. The
accuracy of time is in milliseconds.
Choice of Experimental Design Selected experimental design is a factorial
design with two levels [1]. The design is shown in figure 1.
The experiment is complete factorial (full resolution). There is no need to
create this design with blocks.
Experimentation The order of runs, one repeat of 2
5
combinations = 32 runs,
are fully randomized using a uniform distribution. The purpose of the random-
ization is to lower the potential effect on load caused by non-simulator processes
on the cluster.
3
Analysis
3.1
Variance Model
The wall clock time T
ijklm
is assumed to be a function of the parameters in the
following way:
MarkovKillers:
i
PlanAgents:
j
94
Performance of the Zereal MMOG Simulator