Sunday, 24 February 2013  


Trade-off availability/consistency formalized based on UML/OCL and metrics

The trade-off availability/consistency has been formalized using UML (Unified Modelling Language) and OCL (Object Constraint Language). Several adaptive replication protocols have been designed that support the trading process. Furthermore, metrics have been identified for different aspects of the trading and guidelines of how to use them have been devised.

System models and architectures for balanced dependability

We have defined two different platform-independent system models for enhanced dependability that comprise a system architecture and interaction and composition of main system elements. One of the models is targeted to tightly coupled, data-centric systems (distributed object or component systems, typically supported by middleware) and the other one is focused on loosely coupled, service-oriented systems (Grid, peer-to-peer, Web services).

Platform-specific mappings and prototype implementations

The platform-independent system models have been mapped to and implemented on different platforms (EJB, CORBA, .NET, and Globus). Furthermore, the metrics defined in the first phase have been used to verify that the system is fulfilling its requirements by carrying out the defined test cases.

Comparison between the system models and recommendations

The technologies of the tightly coupled track have been compared and assessed in the context of industry scenarios. This comparison basically shows more commonalities than differences. The major differences between the technologies are in the set of application programming interfaces (APIs) and libraries freely and/or commercially available to be used. Moreover, the complexity of the reconciliation phase in general has been identified as the most critical factor to decide, whether the DeDiSys approach is applicable or not.

In addition, the tightly and loosely coupled tracks have been compared with each other on the conceptual level. Generally, while the application scenarios of tightly-coupled applications are very specific, the DeDiSys framework will provide most benefits to loosely-coupled applications deployed in large, wide-area inter-organizational networks, where the environment can not be controlled and the ability to make a priori assumptions about the environment during development is severely limited, and other approaches prove impossible or prohibitively expensive.