Peer-To-Peer Architectures in Distributed Data Management Systems for Large Hadron Collider Experiments

The main goal of the presented research is to investigate Peer-to-Peer architectures and to leverage distributed services to support networked autonomous systems. The research work focuses on development and demonstration of technologies suitable for providing autonomy and flexibility in the context of distributed network management and distributed data acquisition. A network management system enables the network administrator to monitor a computer network and properly handle any failure that can arise within the network. An online data acquisition (DAQ) system for high-energy physics experiments has to collect, combine, filter, and store for later analysis a huge amount of data, describing subatomic particles collision events. Both domains have tight constraints which are discussed and tackled in this work. New emerging paradigms have been investigated to design novel middleware architectures for such distributed systems, particularly the Active Networks paradigm and the Peer-to-Peer paradigm. A network management framework has been designed and developed, which is able to carry on network management tasks and failures detection. Moreover an Artificial Intelligence based autonomous agent has been prototyped, in order to support the network administrator in pursuing his management responsibilities. The Active Networks paradigm has been used in this context to enable network programming. In addition, a software prototype has been developed to enable controlling of DAQ systems by means of distributed discovery services. A discovery service allows network entities to be acknowledged about each other, and enables them to expose and make use of custom services. The Peer-to-Peer paradigm has been leveraged to implement a discovery service to ease configuration and monitoring of distributed DAQ systems. The research and development activity carried on in both domains had the common goal of demonstrating how appropriate middleware can provide the required autonomy to systems under analysis.