Information Technology Reference
In-Depth Information
task consisting of 100,000 partial tasks, and block graphs
B
1,
B
2 and
B
33 show a task
consisting of 1,000,000 partial tasks. The research was carried out for the following
variants of the operation of the agent system (Fig.
4.6
):
•
A group of block graphs
A
1,
B
1 presents the efficiency of the calculation of tasks
when the agents
A
g
1
,
A
g
2
,
A
g
3
did not operate in the system, and sending tasks
only between the neighbouring processors was used for balancing the distribution
of tasks.
•
A group of block graphs
A
2,
B
2 presents the efficiency of the calculation of tasks
when only the agents
A
g
1
,
A
g
2
acted in the system.
•
A group of block graphs
A
3,
B
3 presents the efficiency of the calculation of tasks
when only the agents of types
A
g
1
,
A
g
2
as well as
A
g
3
—so-called unemployed
agents acted in the system.
We may observe that if the number of calculated tasks is larger, the saturation of the
structure with tasks takes place and irregularities of their distribution are smaller,
which makes the balancing of the task distribution easier.
Figure
4.7
presents the momentary values of the indicator
W
q
for the calculation of
the task consisting of 100,000 partial tasks in three different variants of the operation
of the agent system:
•
Figure
4.7
a presents the indicator
W
q
for the calculation of tasks when the agents
A
g
1
,
A
g
2
,
A
g
3
did not operate in the system and sending tasks only between the
neighbouring processors was used.
•
Figure
4.7
b presents the indicator
W
q
for the calculation of tasks when only
the agents
A
g
1
,
A
g
2
acted in the system.
•
Figure
4.7
c presents the indicator
W
q
for the calculation of tasks when only the
agents Ag1, Ag2 as well as
A
g
3
acted in the system.
The greater irregularity in distribution in the system, the higher the value of the
indicator
W
q
is. We may observe that at the beginning of calculations there is a
certain irregularity but it later disappears. It results from two factors: the increasing
number of generated partial tasks (saturation) and the beginning of the sending system
operation also with the use of the agent system (variants b and c).
However, at the end of the calculations the number of tasks decreases (generation
is blocked) and saturation begins to fall, and then an underflow of tasks appears in
the system. At that time, the role of the agent system becomes significant due to
which the period of appearing irregularity becomes shorter, hence the calculation of
the whole task is faster.
Figure
4.8
presents the changes of the number of agents in the system at the time
of calculations of the task consisting of 1,000,000 partial tasks. In this graph,
NA
g
1
denotes the number of the agents
A
g
1
,
NA
g
2
—the number of the agents
A
g
2
, and
NA
g
3
—the number of the agents
A
g
3
. The graph
NA
g
presents the cumulative
number of the agents
A
g
1
,
A
g
2
and
A
g
3 acting in the system at a given moment.
Analyzing these graphs, we may conclude that at the beginning of calculations
a momentary increase in the agents
A
g
1
searching for tasks appears. However, it
