Regatta: a framework for automated supervision of network clouds

Abstract

As network technology has become more powerful and complex, it has become increasingly difficult to monitor the well-being of a network. Failure semantics, media, network equipment, and topologies are quickly evolving towards greater sophistication and lesser serviceability. We often wish that the locus of our network supervision were inside the network, as close as possible to the suspect faulty equipment, in order to perform unit verification and real-time in situ repairs of network equipment. Unlike simple end-to-end pings, this directional, sub-end-to-end supervision requires a wide range of data concerning topologies, type of network nodes, node contingencies, and proven test suites for them. Our framework makes supervision of network nodes (for example, switches, routers, firewalls) and node aggregates (clouds) practical by automating the acquisition and utilization of this data. The resulting platform marries the ease of end-to-end monitoring (the operator does not know about intervening network nodes or topologies) with the merits of unit supervision into the network (highly specialized coverage). Regatta overcomes key weaknesses in both end-to-end monitoring (wherein a whole end-to-end extent is the sole testing scope of choice) and unit supervision into the network (that is, multi-dimensional dependencies on the network itself). A Regatta's limiting factor is in the availability of Java runtimes inside the network - such runtimes are required in order to dynamically inflate and customize Regatta's artifacts. Short of active networks or mobile agents infrastructures supporting Regatta's widespread application, the Regatta framework already yields an appealing supervision resolution when attached to actual networks, and scales up to the finest resolution in experimental settings with all Java-enabled network nodes.