ActorOriented Reconfiguration Stephen Neuendorffer Reconfiguration Ports Key Idea

Actor-Oriented Reconfiguration Stephen Neuendorffer Reconfiguration Ports Key Idea: reconfiguration allowed only at quiescent points of the actor performing the reconfiguration. A Semantic Menagerie Default value if no reconfiguration. PN PSDF HDF SDF An actor that can reconfigure a parameter P, or any parameter that P depends on, is a change context of P. More Reconfigurable A single value is consumed from reconfiguration port before other inputs are processed. More Analyzable Data values received from a reconfiguration port set the value of a parameter with the same name. Hierarchical models are trees. A B C Version 4. 0 includes anew behavioral type system that can verify a wide variety of constraints on reconfiguration. This type system represents a unified mechanism for analyzing reconfiguration. ) = D C A QC µ QA Approximate the set of actors that are change contexts for P by the greatest lower bound of the set. > A B C ? Artificial Top element represents the GLB of the empty set. i. e. a parameter with no change contexts Artificial Bottom element ensures that the GLB exists. bpc = u fchange contexts of pg > b c D If p 2 depends on p 1 , then p 1 ? p 2 > If actor c recon¯gures p, then c D? bpc Placeholder Models Source of a mobile actor. “A contains B” “A contains C” “A contains itself” The set of quiescent points of an actor contains the set of quiescent points of any model the actor is contained by. Modal Models Given different syntaxes for reconfiguration and different constraints on reconfiguration, what is good actor-oriented infrastructure for analyzing reconfiguration? ADB ADC ADA If p is a parameter that determines the behavior of > an actor a, then bpc D? a “Placeholder” component allows substitution of different controllers If p is a rate parameter used to schedule > > b c D an SDF model m, then p ? If p is a rate parameter used to schedule > b c D an Parameterized SDF model m, then p ? m Mobile actor is transmitted over a network and installed dynamically at run time. P hase. State. Estimator b c D> ? P hase. State. Estimator: input. Count b c D> ? Compute. H istogram: input: token. Consumption. Rate b c D> ? Compute. H istogram: SDF schedule D> ? P hase. State. Estimator The least change context approximation discards no interesting semantic information! The solution to these constraints can be found in linear time.