21th NMRG Meeting, statement for future research directions in network management from NEC - Network Laboratories, Heidelberg Giorgio Nunzi Distributing management has been one of the dominating paradigms in the research community during the past decade. Autonomic computing showed up as a new reference architecture in the last years and self-x technologies have been investigated in different forms. Interests are now shifting towards the technology transfer of these results to products: - Mesh networks have clear deployment scenarios and have been realized as pilot tests or operative networks. - The Beyond 3G effort finally defined the new architecture for mobile telecommunication networks: it is clear that it is based on a distributed paradigm and many management functions will be delegated down to the base stations. - Plug-and-play and adaptation are technology values added to many products to attract operators' interests for costs savings in installation and maintenance. These examples shows that self-x technology are receiving interest and certainly will be transferred into future network solutions. In this scenario, the future challenging questions to address will be: Can the transfer of self-x technology be as easy as a pure adoption? Can normal management processes be executed on top of the new architecture? The problem arises from the conviction that many management functions will still be maintained with centralized processes. Depending of the application scenario, this will probably concern fault management, goals enforcement, accounting, etc. Therefore, a bridge must be built between the two worlds, the population of self-managed functions and the centralized functions. The instruments required to do this can vary from correlation between different nodes, to filtering autonomous functions and aggregation of information. The questions to pose to the research community are: How can we extract information from self-managed nodes? How can we correlate differences between them? How do we present management information to human operators in an easy way? Traditional instruments might not be adequate and might require further study. It is unfeasible to put classical management interfaces on top of autonomic functions. Data mining techniques might not reveal efficient to discover a behaviour over a distributed self-managed network. The co-existence of different management interfaces is likely to be accentuated with the use of self-x functionalities. On top, the human interface needs to be revised as well to cope with all the possible approaches used in network management. A challenging direction is towards an integrated instrument to cope with different information elements, types of interface, correlation policies and methods to enforce decisions. Will the winning approach in network management rely on a "Google-like" approach, integrating self-x technologies with central management systems?