Page 114 - DCAP608_REAL TIME SYSTEMS
P. 114

Unit 11: Application of Clock-driven Scheduling




          source actor of e, cns(e), a positive integer that represents the number of tokens consumed from  Notes
          e by each firing of the sink actor of e. These numbers are also called the delay, production rate
          and consumption rate, respectively. The source actor and sink actor of e are denoted as src(e) and
          snk(e), respectively. If prd(e) = cns(e) = 1 for each e 2 E, then we say that G is a homogeneous
          SDFG (HSDFG).




             Notes  A special class of algorithms can be modelled by a synchronous data flow (SDF)
             graph for which efficient implementation methods exist.
          The edge e with source actor u and sink actor v is denoted by e = hu; vi. The set of incoming edges
          to actor v is denoted by InE (v), and the set of outgoing edges from v by OutE(v). An initial delay
          distribution of the SDFG G is a vector containing delays on all edges of G, denoted as d(G). An
          SDFG G is sample rate consistent if and only if there exists a positive integer vector q(V) such
          that for each edge e in G,

                                       e
                                   ( q src ( )) prd ( )  ( q snk ( )) cns ( ).
                                                      e
                                              e
                                                             e
          where (1) is called a balance equation. The smallest q is called the repetition vector. We use q to
          represent the repetition vector directly.
                 Example: A balance equation can be constructed for each edge of G1 in Fig. 11.1 (a). By
          solving these balance equations, we have G1’s repetition vector q = [2, 1, 1].
          One iteration of an SDFG G is a firing sequence in which each actor v occurs exactly q(v) times.


                 Example: An iteration of G1 in Fig. 11.1 (a), for example, includes two firings of actor A
          and one firing of B and C, respectively.

          Self Assessment

          Fill in the blanks:
          5.   ……………… are usually required to operate under real-time constraints and with limited
               resources.
          6.   The construction of rate-optimal schedules involves ……………… retiming and unfolding.
          7.   It is always possible to convert an SDFG to its equivalent ……………… and then use the
               available methods for HSDFGs.
          8.   The average computation time per iteration is called the ……………… of a schedule.
          9.   A ……………… arranges computations of an algorithm to be executed repeatedly.

          11.3 Pros and Cons of Clock-Driven Scheduling

          The advantages of clock-driven scheduling are discussed below:

              Conceptual simplicity
                   complex  dependencies,  communication  delays,  resource  contentions  can  be
                    considered when developing the schedule






                                           LOVELY PROFESSIONAL UNIVERSITY                                   109
   109   110   111   112   113   114   115   116   117   118   119