Database Reference
In-Depth Information
Fig. 8.4
Aggregation functions for each QoS attribute
a specified VM. Same application service has different execution QoS in different
VMs.
Network QoS
refers to the QoS for transferring data from one application
service to another using a specified network UCS. Data transfers are determined by
the source services and the destination services. Each data will be transferred as soon
as the source service produces them. Hence, network QoS values are only calculated
at the destination services.
Storage QoS
refers to the QoS for storing certain amount
of data for a certain time using specified database service. Assume no data will be
stored during the execution of an application service. Therefore, the only data needs
to be stored are the input data. For example, a destination service has two input
data. One input data arrives early, the other arrives later. The earlier arrived data
need to be stored when waiting the second input data to arrive. The QoS value for
a service therefore equals to the sum of execution QoS, network QoS and storage
QoS. Figure
8.4
shows the aggregation functions for calculating the overall QoS
for composite services. m is the number of component services in the composite
service. QoS values are normalized using Simple Additive Weighting (SAW), which
is also used in [
238
]. The best QoS values are normalized to 0, the worst QoS values
are normalized to 1. Thus, higher normalized values indicate worse quality.
QoS constraints (denoted as QC ) for composite services have two types:
Global
Constraints
and
Local Constraints
. Global Constraints are the QoS constraints for
the overall composite service, while Local Constraints apply to component services
within the composition. A global constraint (GC) for a given QoS attribute Q
l
is
denoted as GC
l
. Local constraints are denoted as LC
l
. Constraints on different
QoS attributes are transformed into inequality constraints [
107
]. QC
1
(time) and
QC
2
(price) can be transformed by subtract the threshold to the constraints, e.g.
QC
1
1min
u
te is transformed to QC
1
(
QC
1
1
0; QC
2
5USdollars
is transformed to QC
2
5
0. QC
3
(availability) and QC
4
(reputation)
can be transformed by subtracting the QoS value from the threshold, e.g. QC
3
(
QC
2
0:9 is transformed to QC
3
(
0:9
QC
3
0. Constraints on equal QoS attributes
can be transformed using this function: QC
(j
QC
j
0, where is the
tolerance allowed range (a very small value).
Genetic Algorithms
Genetic Algorithms
(GAs) are heuristic approaches to iteratively find near-optimal
solutions in large search spaces. Any possible solution to the optimization problem
is encoded as a
Chromosome
(normally a string). A set of chromosomes is referred
