Minimizing buffer requirements for throughput constrained parallel execution of synchronous dataflow graph

Abstract

This paper concerns throughput-constrained parallel execution of synchronous data flow graphs. This paper assumes static mapping and dynamic scheduling of nodes, which has several benefits over static scheduling approaches. We determine the buffer size of all arcs to minimize the total buffer size while satisfying a throughput constraint. Dynamic scheduling is able to achieve the similar throughput performance as the static scheduling does by unfolding the given SDF graph. A key issue of dynamic scheduling is how to assign the priority to each node invocation, which is also discussed in this paper. Since the problem is NP-hard, we present a heuristic based on a genetic algorithm. The experimental results confirm the viability of the proposed technique.

Resource minimized static mapping and dynamic scheduling of SDF graphs

“…In this way we can build the static scheduling result which concerns graph unfolding. The last set of experiments compares three priority assignment methods: method used in [7], invocation-based assignment, and node-based assignment explained in the previous section. We vary the throughput constraint to find a mapping that requires the minimum resource overhead.…”

“…For mapping/scheduling optimization, previous works are all based on static scheduling. There is an exception in [7], which assumes that a static mapping decision is given from any previous work and proposes a technique to minimize the buffer requirement assuming that a dynamic scheduling technique is used. They did not consider how to determine the static mapping.…”

“…In addition to mapping and buffer size information, we encode priority information for dynamic scheduling in a single composite chromosome together. In addition to a simple heuristic approach used in [7], two methods are explored in our design flow: invocation based and node based method. The invocation based method is to assign a priority value to each node invocation freely.…”

“…As a baseline method, we assign the priorities first according to the method used in [7] and find a sub-optimal mapping for dynamic scheduling next. This method is based on the critical path length as used in Figure 2 (c).…”

“…Our previous work [7] finds the minimum buffer requirements to meet a given throughput constraint when dynamic scheduling is utilized assuming that the mapping decision is given under the assumption of the static-assignment strategy. The problem addressed in this paper is, first, to find a static mapping that minimizes the resource requirement with satisfying the throughput.…”