7.0 Conclusions

be able to obtain a gain in performance, essentially for free, by performing the post-processing step suggested in Claim 1. What we have proposed can therefore be added as the final step in existing schedulers. Of course, an exhaustive search procedure like the one proposed in [11] will certainly find the schedule directly. For an ordered transaction strategy, we use the transaction ordering suggested by the modified schedule rather than the transaction order from used in Fig. 5. Thus imposing the transaction order 2→1, 4→3, 5→2, 1→4, 3→2, and 1→5 as in Fig. 6 results in of 9 units instead of 10 that one gets if the transaction order of Fig. 4 is used. Under the transaction order specified by , ; thus this order ensures that the average period is within one unit of the unconstrained self-timed strategy. Again, unfolding may be required to obtain a transaction ordered schedule that has period exactly equal to , but the extra cost of a larger controller (to enforce the transaction ordering) outweighs the small gain of at most one unit reduction in the iteration period. Thus for all practical purposes the transaction order specified by is the optimal order. The " optimality " is of course only under the assumption that the specified execution times of actors are accurate, and under the constraint that the processor assignment and order of execution of actors is kept the same as the original fully static schedule. In other words the transaction order we determine is the best possible one for the available timing information, given the processor assignment and actor ordering. If the generated fully-static schedule is to be run in a self-timed fashion, then of course there is no need for the post-processing step of Claim 1. Determining the order of processor transactions at compile time and enforcing this order at run time leads to a low-cost inter-processor communication mechanism. In this paper we have shown how to determine the best possible transaction order under the given timing information. The procedure, instead of simply extracting the transaction order from the given fully-static schedule, first modifies the fully-static schedule by skewing the starting times of processors. The resulting fully-static schedule has a period within one time unit of the average period obtained if the same schedule were run in a self-timed fashion. Using the transaction ordering specified by the modified schedule S′ S′ S …