Worst-case performance analysis of synchronous dataflow scenarios

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 worstcase 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 optim...

• Type: Journal
• Year: 2010
• Where: CODES
• Authors: Marc Geilen, Sander Stuijk
• Keywords: CODES 2010, Dataflow, Dataflow Model, Scenario, Software Engineering