Providing Decentralised Adaptive Services — Entwicklung und Betrieb dezentraler adaptiver Dienste

Recently, many research activities have investigated means to cope with the exponential growth of computing power, numbers of devices, and most importantly, information. The main focus of these activities is managing the complexity and optimising resource utilisation. This thesis presents novel solutions that optimise the utilisation of distributed resources for a stable and scalable execution of long-running services. This challenge is tackled by providing development and infrastructure support for two object-oriented distributed service models named replicated decentralised adaptive services and peer-to-peer based decentralised adaptive services. Both service models avoid a central point of failure and are capable of adapting their distribution according to service and environment demands. Replicated decentralised adaptive services are supported by a framework that enables the flexible development of fault-tolerant services and supports active replication. In contrary to related approaches the proposed solution builds on top of the fragmented object model as provided by the Aspectix middleware. This approach is a lean way of adding faulttolerant replication to services based on standard middleware. Furthermore, the framework retains the flexibility of the fragmented object paradigm by providing semantic annotation support at the interface level to enable a custom code generation for an optimised handling of requests and the introduction of object-specific code at client-side. This flexibility enables the development of complex dependable distributed services as this thesis demonstrates by the implementation of a distributed version control system. Dynamic distribution of replicated decentralised adaptive service is enabled by an extended reference concept that ensures the reliable binding of services despite location changes. Furthermore, support for a fast state transfer of arbitrarily large service states is provided. Peer-to-peer based decentralised adaptive services target the utilisation of distributed resources using peer-to-peer concepts in context of a standard middleware. The proposed solution integrates the JXTA peer-to-peer infrastructure into CORBA using the fragmented object support of Aspectix. Services can make use of peer-to-peer mechanisms to be more scalable and fault-tolerant than traditional client/server distributed objects. Existing CORBA applications can now transparently access such extended services. Both service models constitute the basis for the environment for decentralised adaptive services (EDAS), which enables the distribution and management of services across multiple administrative domains. As dynamic distribution might move services to nodes on which the required service-specific code is not available, EDAS offers support for dynamic loading of code. The presented loading service differs from related approaches, as it uses a decentralised peer-to-peer based lookup, selection and loading process for providing platformspecific code. Finally, EDAS advances the state of the art in distributed resource management. The supported mechanisms include a novel decentralised and fault-tolerant protocol to enforce global resource quotas and a decentralised placement algorithm that enables a resourceaware distribution of decentralised adaptive services.