Database Reference
In-Depth Information
Fig. 8.7
Example of
scheduling string. (
a
)
Example 1. (
b
) Example 2
for data forwarding
S
2
starts to execute just after S
1
starts, S
1
and S
2
can be considered to start their
execution at the same time. If
data
2
is available (S
2
executes faster) before
data
1
,
data
2
will be stored in S
2
's database service till
data
1
has been transferred to S
3
.
2. Another important feature is
data forwarding
[
229
]. For an input data, the source
service can be chosen among the services that produce or consume this input
data. All the consumers of this input data can be forwarders. For example,
Fig.
8.7
b shows a scheduling string. S
2
and S
3
both have the input data from S
1
.
S
2
may forward
data
1
from S
1
to S
3
, i.e. shown as the dashed line in Fig.
8.7
b.
This kind of data forwarding is not allowed in our work. Data must be only
transferred from the original data producer to its consumers.
Stop criterions for the proposed approach are: (1) Iterate until the constraints
are met (i.e. D.c/
D
0). (2) If this does not happen within MAXGEN
generations, then iterate until the best fitness value remains unchanged for a
given number (MAXGEN ) of generations. (3) If neither (1) nor (2) happens
within MAXGEN generations, then no solution will be returned.
Handling Multiple Data Flow Graphs
Assume the composite service (e.g. shown in Fig.
8.2
) has multiple data-flow graphs
(shown in Fig.
8.3
). For each data-flow graph, an optimal composition solution can
be generated using the proposed GA-based approach. Since each of the optimal
solution only covers a subset of the composite service, further actions are needed
to aggregate these partial composition solutions into an overall solution. Assume
the composite service has f data-flow graphs (i.e. df g
1
;dfg
2
;:::;dfg
f
). The
approach adopts the following strategies to aggregate multiple solutions into an
overall solution:
Given an abstract service S
i
,ifS
i
only belongs to one data-flow graph (e.g.
df g
j
), then the proposed approach selects df g
j
's solution
chromosome
j
to
execute abstract service S
i
.
