发表于Data & Knowledge Engineering 67 (2008) 463–484 #2009science
服务组合分为两种类型：orchestrated services和choreographed services 集中式和分布式
Complex services can be handled both as orchestrated services, and thus executed by an orchestration engine or, under a more loosely-coupled perspective, as choreographed services, expressing constraints on admissible ordering of messages [30]. A choreography specifies the cooperation among different actors, each following a globally defined interaction protocol, but with no central execution engine, run by one of the participants.
2-1-4QoS prediction
In our approach, QoS faults can be predicted through a prediction model and a support framework based on monitoring and machine learning.
（1）Some regularities can emerge from the observation of the Global QoS, depending on the values of specific QoS parameters observed in different situations.
（2）This objective can be achieved by using Machine Learning (ML) techniques [38] to provide suggestions on possible variations of the Global QoS in a run-time environment.
QoS monitoring model
目的：
The monitoring functionality cover different purposes: (i) to check if the execution of the complex service correctly follows the interaction protocol defined by the global choreography, (ii) to check if a QoS contract is respected, (iii) to estimate the QoS of next operations for preventing QoS fault.
实现方式：
More precisely, the monitoring functionality is implemented by two logical components:
- A central Choreography Monitor, implemented as a Web-Service, that uses the description of the global choreography to check the complex service execution, and that interacts with the services involved in the choreography through the Message Handlers (see Fig. 7).
- Inbound/Outbound Message Handlers, implemented by all the Web-Services participating in the choreography, that (i) intercept inbound/outbound messages and notify their arrival/departure to the Monitor; (ii) expose a communication interface used by the Monitor to query services about their status
QoS prediction system
The way the QoS prediction system builds the prediction is described in Section 2.1.4. In details, the Web-Service has two interfaces, namely:
(1) getClassificator(TrainingSet), that builds a classifier from the given training set.
(2) getClassificationResults(Instance), that returns a prediction over the given Instance using the classifier built in the previous step.
4. Reaction and adaptation to QoS faults
The recovery functionality can be activated to react to a Choreography fault, a QoS fault or a prediction of a QoS fault.
