Decentralized Adaptive Power Control for Process Networks
(English)Manuscript (preprint) (Other academic)
Mobility and location transparency of distributed objects enable efficient resource allocation in networks, and can be effectively realized at the language level using location independent routing. However, the benefits of mobility are restricted by the available resources, e.g., in the form of processing nodes, which may be too few or too many to suit an application. To continually meet an application developer's requirements on computational task throughput, and an infrastructure provider's requirements on energy usage, the network itself must be able to adapt. In this paper, we consider fault-free networks of nodes connected point-to-point by asynchronous message passing channels, and propose a protocol for shutdown of nodes that preserves the integrity of distributed objects. This protocol enables decentralized power control, where nodes are turned on and off in response to computational requirements. We analyze the protocol both by verifying a restricted version in the model checker Spin, and by formulating a transition system model and proving properties by induction in that model for networks of arbitrary size. We define a protocol extension that, while using only a node's neighborhood-local information, is sufficient to ensure networks remain connected after node shutdown, and outline more complex, general local criteria. Finally, we discuss heuristics for node-local decision making on initiating a shutdown process, to meet adaptability requirements.
power control, object mobility, network protocols, adaptability
Research subject Computer Science
IdentifiersURN: urn:nbn:se:kth:diva-152240OAI: oai:DiVA.org:kth-152240DiVA: diva2:749491
QS 20142014-09-242014-09-242014-09-29Bibliographically approved