Ebros or how to speed date your network

Enterprise data centres are challenging environments for the configuration, troubleshooting and operational management of the network. Every day network administrators, despite automation efforts, still need to make adhoc management decisions. In this they often rely on long-term experience and intimate understanding of low-level system specifics. In this context, a platform that captures operators'

routine knowledge as well as structural and operational features of the network is a natural match to the needs of network management.

Ebros is our vision of one such platform. Ebros is a formalized model of the network that has the ambition to capture the network in full detail and to support query-based interaction with human operators.

The model can be seen as a labelled property graph, in which vertices represent network elements and edges - the connections between them.

In contrast to earlier works on graph representations for networks, we do not limit the representation context. Instead, Ebros targets all layers of the network stack and the relations between the layers. Our design approach results in a vast model base; to streamline the interaction with the model we provide mechanisms for the generation of abstractions or logical views. In more informal terms, Ebros can be seen as a platform for network operators to speed date the network asking only the questions that interest them and getting to know the answers.

Desislava Dimitrova is a Postdoctoral researcher at the Systems Group at ETH Zurich. Her focus is on the analysis and modelling of network behaviour in modern data centres. To this end she explores the use of data representation techniques, distribute processing platforms and programmable network concepts to address current challenges in data centre network management. Prior to her current position she was involved in research on virtualization and resource management strategies for mobile networks and the use of wireless signals for indoor localization. Her broader interests include sensor networks, smart environments, and workload characterization.