A reliability model for distributed adaptation

End-to-end connectivity is growing increasingly diverse, with orders-of-magnitude differences in characteristics throughout the network. At the same time, most applications assume a level of network characteristics below which they either provide no service, or service at a cost that is higher than the user is willing to pay. Open-architecture networks can help applications degrade gracefully when network conditions are poor by pushing adaptation technology into the network. Unfortunately, since adaptation seeks to change the data stream during transmission, it is incompatible with the traditional model for reliable data streams. A new model of reliability is required to allow general the adaptation of reliable data streams. We propose one possible model that provides reliable delivery of the semantic meaning in the data stream, despite adaptation. We present this model in the context of Conductor, a framework for distributed adaptation. By using a reliability model that is compatible with adaptation, Conductor allows arbitrary distributed components to operate on a data stream without reducing end-to-end reliability. While Conductor is one possible adaptation service, it is also an example of the type of reliability model required in open-architecture networks