Elastic network control: An alternative to active networks

Much of the recent work in network control has concentrated on opening up the network, moving away from closed, monolithic control systems. Such work can be placed into two categories: use of generic APIs, giving rise to open control of networks; and allowing packets on the datapath to program the network nodes, leading to the concept of the active network. Drawing on experience in both of these areas, this paper presents the design and implementation of a framework that enables clients to program all aspects of the network in a manageable, controllable manner. By using resource partitioning, clients are allowed to inject code into the various entities of the network's control system, subject to access control restraints. This form of programmable network is termed an elastic network and it is proposed as an alternative to current active networking approaches. The elastic network has two key advantages over open control and active networks. Firstly, the elastic network allows dynamic customization of all aspects of the network control, management and data processing components. Secondly, all customization takes place in a safe, resourcecontrolled way. This makes available the flexibility of active networks, and the resource controllability of open control networks, without imposing the attendant costs on all users of the network. Costs still exist, but their imposition is restricted to those places where the available functionality is actually used.