Performance analysis of weakly-consistent scenario-aware dataflow graphs

Abstract

The timed dataflow model of computation is a useful performance analysis tool for Electronic System Level Design automation and embedded software synthesis. It is used to model systems, including platform mapping and resource scheduling, of components communicating and synchronizing in regular patterns. Its determinism gives it strong analysability properties and makes it less subject to state-space explosion problems. Because of its monotonic temporal behaviour it can provide hard real-time guarantees on throughput and latency. It is expressive enough to cover a fairly large class of applications and platforms. The trend however, in both embedded applications and their platforms is to become more dynamic, reaching the limits of what the model can express and analyse with tight performance guarantees. Scenario-aware dataflow (SADF) is an extension that allows more dynamism to be expressed, introducing a controlled amount of non-determinism into the model to represent different scenarios of behaviour. The combination of a relatively infrequent switching between scenarios and still deterministic dataflow behaviour within scenarios stretches the expressiveness of the model while keeping sufficient analysability. In this report we investigate so-called weakly consistent graphs in which the scenario changes are not tightly coupled with periods of repetitive behaviour of the static dataflow behaviour in scenarios as in previous methods. We define their semantics in terms of (max, +) algebra and we introduce a method to analyse throughput using a generalisation of (max, +)-automata.

Scalable Analysis for Multi-Scale Dataflow Models

“…Since transitions are at the level of individual firings of actors, the resulting state space becomes very large with larger models and leads to tractability issues. The FSM-SADF analysis [11,13] uses a symbolic simulation method to get the (max, +) representation of each scenario. This representation abstracts the actor firing dependencies within the scenarios.…”

“…Irrespective of the size of the resulting directed graph or the automaton and the complexity of the final MCR analysis, the process of generating the state space of the time-stamp vectors or the automaton scales linearly in the number of FSM states. We use the same symbolic simulation method to obtain (max,+) representations for scenarios as Geilen et al [12,13]. Then we use a transformation on the given FSM-SADF to generate a new FSM-SADF with fewer number of FSM states, leading to further scalability improvements for multi-scale dataflow models compared to the analyses of Geilen et al [12,13].…”

“…This section provides a formal definition of FSM-SADF and defines the performance metrics we provide analysis for. The definition of FSM-SADF is adapted from the definition provided by Geilen et al [11]. An FSM-SADF is defined by a tuple (S, д, r, i, f , o, L)…”

“…A throughput analysis based on this definition is provided by Geilen et al [11]. The throughput in the example of Fig.…”

“…Without losing accuracy, the scenarios of the resultant FSM-SADF are defined to abstract these sets of scenario sequences and a new regular expression is constructed using the defined scenarios. The final analysis invokes the existing FSM-SADF analysis [11] with a compact FSM that recognizes the language formalized by the new expression. We prove that the resultant FSM-SADF has the same worst-case throughput as the original FSM-SADF.…”