110 likes | 123 Views
This paper introduces the concept of endochronous systems and presents the Kahn principle for deterministic networks of such systems. It outlines the proof of the theorem and discusses practical consequences and future work.
E N D
The Kahn Principle for Networks of Endochronous Programs Dumitru Potop-Butucaru IRISA, France
Outline • Introduction • Motivation • Endo/Isochrony • Endochronous systems • The Kahn principle • Proof outline • Conclusion. Future work
Introduction • Synchronous systems, composition • Desynchronization: • Goals: Input reading from async. sources Semantics preservation: (Σ1| Σ2)a= Σ1a || Σ2a Σ Σa
Endo/Isochrony (Benveniste et al.) • Endochrony (of a system) • Do not use input signal presence information • Isochrony (of a pair of systems) • Non conflicting transitions are composable Theorem: a. Isochrony semantics preservation b. Endochrony semantics preservation Isochrony • Non-deterministic, no intra-instant causality
Endochronous system model Σ=(S,V=INOUT,O,θ,) Microstep states: M ={(s,I,φ)|sS,IIN, φ:I D } Output functions: O:M P(OUT) ; Θ(μ):O(μ) D Input decision function, monotonous: READ:MP (IN), IREAD(s,I,φ) Transitions: (s,I,φ)(s’,I’,φ’): intra-instant: s=s’ and II’READ(s,I,φ) and φφ’ state change: READ(s,I,φ)=I and I’= and φ’=
Endochronous system model • Incrementally read the inputs from input channels (e.g. unbounded FIFOs) • The decision to read a signal is never revoked (s,{i1},{(i1,v1)}),READ={i1,i2} (s,{i1,i2},{(i1,v1),(i2,v2)}),READ={i1,i2} (s,,),READ={i1,i2} (s,{i2},{(i2,v2)}),READ={i1,i2}
Asynchronous network model • Network: Synchronous systems + FIFOs • FIFO state: words over D • Network state: component, FIFO states • Network transition: component trans. • Possible if the FIFOs to read are not empty • Change the status of the component and the status of some adjacent FIFOs
The Kahn Principle Theorem Asynchronous networks of endochronous systems are deterministic as I/O functions • Inter-component concurrency does not affect the semantics of the system • The result holds for bounded FIFOs • Isochrony 1-place FIFOs are enough
Proof outline • Step 1: I/O determinism of a component for an instant (monotony of READ) • Step 2: I/O determinism of a component (step 1 + state change determinism) • Step 3: determinism of the network (Finite derivations, from step 2. Infinite, by continuity.)
Practical consequences • Make systems endochronous (Signal compiler, add “clock” signals) • When testing, varying the transmission delays (and jitter) of the FIFOs does not influence the outcome of the computation • only one run is necessary for a given asynchronous history
Conclusion • Endochronous component model with causality • Asynchronous system model • The Kahn principle • Future work: extend endochrony to allow internal concurrency