Database Reference
In-Depth Information
Every approach runs 50 times for each scenario. All the results shown below
are the average values from these experiments. Each experiment for the GA-based
approach starts from a different initial population each time. The probability of
crossover p
cross
D
0:4 is the same for the matching string and scheduling string.
The probability of mutation p
mut
D
0:1 is also the same for the matching string
and scheduling string. The approach uses rank-based roulette wheel schema for
selection. The angle ratio of the sectors on the roulette wheel for two adjacently
ranked chromosomes, i.e. R, was chosen to be 1
C
1=P , where P is the population
size. By using this simple formula, the angle ratio between the slots of the best and
median chromosomes for P
D
50 (and also for P
D
200 for large-scale scenarios)
is very closely to the optimal empirical ratio value of 1:5 in [
231
]. MAXFI T
equals to 150. MAXGEN equals to 1; 000.
Exhaustive search approach
would
traverse all the possible solutions to the composition problem and find the optimized
solution that has the smallest fitness value. Although this approach would always
find the most optimal composition solution, the execution time is extremely high.
Random selection approach
is also a GA-based approach. This approach would
randomly select chromosomes to form a new generation. Comparisons with these
approaches show the effectiveness and efficiency of the proposed approach. Integer
Programming (IP) approaches have been proposed to solve QoS-aware service
composition in SOC. The IP approach is implemented using LPSolve [
77
], which is
an open source integer programming system. Comparisons with IP approach show
the scalability of the proposed approach.
Experiments Results
Small-scale experiments are conducted on 10 different test datasets. We only show
two of them in Fig.
8.9
to make the graph much easier to read. Figure
8.8
shows
the results between the proposed approach and the exhaustive search approach.
Proposed GA-based approach would always find near-optimal solution compared
to exhaustive search algorithms. Figure
8.9
shows the comparisons between the
proposed approach and the random selection solution. As shown in this figure,
proposed approach will always reach an optimized fitness value while random
selection seldom converges. To sum up, the proposed GA based approach will
always reach an optimal fitness value and the converged point is very close to the
actual optimal point. Figure
8.10
shows the efficiency of the proposed approach.
These experiments are conducted on small-scale scenarios. Each test dataset has
the same configuration, except for the number of concrete services for each abstract
service. As shown in Fig.
8.10
, the execution time increases quickly at the beginning,
but keeps stable when the number of concrete services for each abstract service is
larger than 200.
As shown in Fig.
8.11
a, IP approach performs as good as the GA based approach
at the beginning. Notice that, when the number of the abstract services becomes
more than 40, IP approaches would cost exponential growing time to solve the
