116 Dynamic Workflow Application Scheduling with Quality of Service
To overcome these limitations, the author proposed to use linear program-
ming of the problem with only real variables (i.e., data flows only) while
adopting a dedicated algorithm for both selection of services and ordering
[60]. For the latter the genetic algorithm can be used. Thus the algorithm
would operate as shown in Listing 4.9. The main loop corresponds to steps of
the genetic algorithm in which a chromosome represents both the selection of
services and their ordering.
The data flow sub-problem can be then solved reasonably quickly as several
constraints can be removed or simplified, having selected services for particular
tasks:
1. only d
ij kl
for the selected services s
ij
and s