SVM-Based Efficient QoS-Aware Runtime Adaptation for Service Oriented Systems
Mingkun Yang, Xiaohui Hu · 2016
Given the increase in web services, Quality of Service (QoS) attributes have been widely addressed. In dynamic Internet environment, service providers rarely deliver the QoS as declared. Runtime adaptations of execution plans become necessary in order to recovery from web service failures, maintain SLAs, and/or continuously improve the overall QoS attributes. As QoS-aware service composition methods are mainly employed to fulfill runtime adaptations, the time efficiency usually becomes a challenge since the execution of a composite application is blocked until the new execution plan is found. Although existing works have provided high-quality methods to deal with QoS-aware service composition at design time, a suitable mechanism has not been established to utilize these methods coping with runtime adaptations in an efficient way. In this paper, we propose a novel empirical approach to accelerate QoS-aware runtime adaptation. Based on historical records, our approach uses Support Vector Machines (SVMs) to capture the relationship between candidate services and adaptation scenarios which are used at runtime to predict the probabilities that candidate services will be used for upcoming adaptation scenarios. Then candidate services are pruned based on these probability estimates to reduce the search space. The experimental results revealed the proposed approach can achieve significantly acceleration while satisfying considerable correctness.