Information Technology Reference
In-Depth Information
2e+06
4
Destination reached
Full GCs
3.5
1.8e+06
3
1.6e+06
2.5
1.4e+06
2
1.2e+06
1.5
1e+06
1
800000
0.5
600000
0
0
200
400
600
800 1000 1200 1400 1600 1800 2000
Launch interval
Fig. 2.
Induced quality and scalability of ant generation
reasonable response time can only be achieved when there is no global garbage
collection at all, we search for the minimum launch interval where global garbage
collection does not occur. We denote this to be the
ideal interval
. In the respec-
tive diagram, this ideal interval length is 500 ms (where the solid graph intersects
the abscissa).
We did similar investigations for other networks having the same number, less
or more nodes. We achieved ideal launch intervals between 100 ms (for 61 nodes)
and 1100 ms for 144 nodes. This meets our expectations that larger networks
require longer launch intervals due to the quadratic increase of (source, target)
pairs. For up to 150 nodes the quality of result (dashed graph) was still nearly
optimum, but for larger networks it declined considerably. Thus, our approach
successfully only applies to networks with up to 150 nodes, at least with the
hardware we used.
For the launch delay, we showed in several tests for different launch intervals
that it is best to keep the delay in a range of 20 % of the launch interval.
4.2 Evaluation of Results and Improving the Algorithm
The goal of our tests was to answer the following questions:
1. How does the system react when a preferred route deteriorates?
2. How fast does AntScout switch back into the initial state when a deteriorated
route has improved again?
3. How does the system react when a so far nonpreferred route improves con-
siderably such that now it would be a preferred candidate?
