A Model of Computation for Real-Time Applications on Embedded Manycores

Transaction Parameterized Dataflow: A Model for Context-Dependent Streaming Applications

“…A sample execution of the graph is the following: A fires and produces p tokens on e 1 . Then B fires and produces one token on edge e 2 , e 3 , e 4 . Only E can fire because there are enough tokens on its input edge and produce one token on edge e 7 .…”

“…In TPDF, we define 2 data distribution kernels Selectduplicate, Transaction Box [4] and a new type of control clock in a dataflow way. a) Select-duplicate: kernels with one entry and n outputs (n is a maximum number for automatic sizing).…”

“…Its role is to atomically select a predefined number of tokens from one or several of its input to its output. By using special modes predefined by TPDF and combining with a control actor, the Transaction process implements important actions not available in usual dataflow MoC: Speculation, Redundancy with vote, Highest priority at a given deadline, Selection of an active data-path among several [4]. c) Clock: can be considered as a watchdog timer with control tokens sent each time there is a timing out.…”

An Empirical Evaluation of a Programming Model for Context-Dependent Real-time Streaming Applications

“…In this context, we have defined a new MoC by extending the CSDF-based usual MoC with 2 important data-filters: Select-duplicate, and Transaction to treat the case of uncertain arrival of data. A more detailed description of these data-filters can be found in section 3 and in [8].…”

“…One challenge, in this context, is any loss of connectivity which can impact parts of its functionalities, but must sustain the core functioning. Another is an increasing requirement for safety and security, even for mass market applications [8]. We must find a way to combine the parallelism management of data-flow principles and the reliability of real-time systems, and a common base to define such a Programming Model.…”