Worst-case performance analysis of Synchronous Dataflow scenarios

Abstract

Synchronous Dataflow (SDF) is a powerful analysis tool for regular, cyclic, parallel task graphs. The behaviour of SDF graphs however is static and therefore not always able to accurately capture the behaviour of modern, dynamic dataflow applications, such as embedded multimedia codecs. An approach to tackle this limitation is by means of scenarios. In this paper we introduce a technique and a tool to automatically analyse a scenario-aware dataflow model for its worst-case performance. A system is specified as a collection of SDF graphs representing individual scenarios of behaviour and a finite state machine that specifies the possible orders of scenario occurrences. This combination accurately captures more dynamic applications and this way provides tighter results than an existing analysis based on a conservative static dataflow model, which is too pessimistic, while looking only at the `worst-case' individual scenario, without considering scenario transitions, can be too optimistic. We introduce a formal semantics of the model, in terms of (max; +) linear system-theory and in particular (max; +) automata. Leveraging existing results and algorithms from this domain, we give throughput analysis and state space generation algorithms for worst-case performance analysis. The method is implemented in a tool and the effectiveness of the approach is experimentally evaluated.