การประกอบเซอร์วิชที่หลายระดับความละเอียดโดยการทำนายเชิงคุณภาพในการให้บริการ / นิธิโรจน์ วงศ์สว่างพานิช = Service composition at various granularity levels with QOS-based prediction / Nitirojht Wongsawangpanich
Service composition is a way of constructing new services from existing services by defining a group of service types or abstract services that are to collaborate, and selecting instances of those abstract services to compose into concrete composite services. This research is interested in two composition issues. (1) Quality of service or QoS is often considered in service instance selection but QoS varies due to conditions and time of use. We cannot consider only the QoS published by service providers when composing services but need to concern about QoS dynamics also. (2) Researches in service composition usually select service instances that match individual abstract services. In the case that the service instances have functional scope that does not match exactly that of the abstract services, they will not be selected. This results in a loss of opportunity to obtain good compositions as those service instances may have good QoS. This research proposes two approaches to these issues. (1) QoS prediction using artificial neural network is proposed for service instances, where QoS here includes response time, reliability, and availability. Predicted QoS values then will be used at different time periods of service composition. (2) Service composition using a genetic algorithm is proposed where service instances with granularity coarser than that of abstract services are also considered. The experiments show that QoS prediction is useful because composition at different time periods yields different composite services. Moreover, considering coarse-grained service instances is also beneficial since they are selected as part of the resulting compositions, and hence the opportunity to obtain composite services with good QoS is increased.