Abstract

and thus must adapt to external conditions; all layers must make cost-based decisions about what mode of operation to use in response to performance feedback. This paper focuses on the generic interface between adjacent layers (client and server) in a multi-level hierarchy. The Adaptive Research Contracts (ARC) framework uses a bottom-up approach in which the server exposes a range of quality/cost modes to the client above. This allows the client to trade off various algorithms generating different workloads for multiple resources. A case study shows that control can be distributed effectively over multiple layers with ARC; global cost-effective solutions can be obtained with exchanging a small fraction of all possible control settings.

Citations