Information Technology Reference
In-Depth Information
service requesters' feedback, and a third party. Its accuracy is
critically important and yet difficult to guarantee.
Web Service Composition under QoS Constraints . Local and
global optimizations are used. Ardagna and Pernici [142] dis-
criminate global and local QoS constraints. The local optimiza-
tion is done at a task level, that is, choosing for each task the Web
service with the best QoS. Casati and Shan [46] propose eFlow to
suit adaptive and dynamic features of Web services required by
different individual users and to cope with a highly dynamic
business environment. To prevent service providers from deviat-
ing from the advertised QoS as such deviation causes losses to the
users, Jurca et al. [143] propose a novel QoS monitoring mecha-
nism to collect the ratings and compute the actual quality deliv-
ered to the users. Huang et al. [144] present a moderated fuzzy
Web service discovery approach to model users' subjective and
fuzzy opinions and to assist service users and providers in reach-
ing a consensus.
Compared with the local optimization approach, the global
optimization is done at the process level where services are selected
for each task to obtain the optimal global quality. Lamparter et al.
[145] use utility function policies to model the multiple preferences
of users. But the utility function is hard to quantify for a multi-
attribute decision. Gao et al. [146] combine different service paths
as a weighted multistage graph and transform the selection of the
optimal execution plan into the selection of the longest path. Since
the service path is defined as a sequential chain of service opera-
tions, there is limitation when the search methods are used in
complex service execution. To overcome the disadvantages, Zeng
et al. [107] use a state chart to describe a workflow execution
sequence in detail, that is, the control-flow, and then arrange a set of
candidate Web services that can equally accomplish the function for
each activity in the state chart. Their work is further improved by
Gao et al. [147] by adding more quality dimensions, that is, capacity
and load, to their QoS model. However, it suffers from several
drawbacks. First, they use 0-1 integer programming [148] to obtain
the solution, which is NP-hard. This chapter proposes the use of a
linear programming method that is solvable in polynomial time
(complexity class P). Second, they assume that every candidate
Search WWH ::




Custom Search