[Back]

O. Moser:
"Holistic Monitoring and Domain Specific Adaptation in Composite Services";
Supervisor, Reviewer: S. Dustdar, F. Leymann; Institut für Informationssysteme, AB Verteilte Systeme, 2012; oral examination: 2012-07-18.

Service orientation emerged as an ubiquitous paradigm to tackle large scale software systems. In this regard, service oriented systems represent application functionalities and responsibilities
by the means of loosely coupled, computational elements, called services. Services can be combined in a structured fashion, resulting in composite services that are used to align enterprise business processes to information technology. Hence, composite services represent a paramount concern in today´s enterprise landscapes. Quality of Service (QoS) provides a means to quantify
certain performance and dependability related metrics that are crucial for the normal operation of composite services, in addition to various business related concerns. Therefore, monitoring
QoS and business related concerns of composite services is an important challenge in service oriented systems. Moreover, such a QoS based quantification can be leveraged to dynamically adapt composite services to changing requirements. Runtime service selection represents one possibility for such dynamic adaptations. QoS can be used to determine slow and unreliable services that affect the overall performance. Consequently, if alternative service instances are available and provide better QoS, a bad performing service can be substituted with a better performing alternative. This thesis addresses the challenges of managing monitoring and adaptation requirements of composite services. Our contributions are threefold. First, we introduce an adaptive QoS model to tackle short term requirements with regard to additional QoS attributes. In this model,
we combine both domain-agnostic and domain-specific QoS attributes. Second, we propose an approach for managing holistic monitoring requirements in composite services. Our approach represents the message exchange within a composite service as a stream of events, and leverages event stream processing technology to address QoS and business related monitoring requirements. Additionally, we treat data that is created outside the scope of a composite service as
first class citizens. This enables us to provide a holistic view on all monitoring data that is available in enterprise environments. Third, we present a method for runtime service selection that
leverages our adaptive QoS model. For specifying the service selection strategy, we propose a domain-specific language. This language can be used to specify selector implementations based
on the available QoS attributes. A genuine case study from the telecommunication domain is used to derive the requirements for such a monitoring and service selection framework. Finally, we provide a proof of concept implementation of our approach for WS-BPEL. This prototype is further used for a performance and qualitative validation of our method.