Throughput Analysis of Synchronous Data Flow Graphs

Analyzing composability of applications on MPSoC platforms

“…Throughput computation algorithms that have a polynomial complexity for HSDFGs, therefore have an exponential complexity for SDFGs. Algorithms have been proposed to reduce average case execution [10], but it still takes in practice O(n 2 ) time where n is the number of vertices in the graph. Besides, even for one application, the number of ways extra edges can be added to model actor dependency is exponential.…”

“…To get an idea of vastness of this number, if there are 5 graphs with 10 actors each we get 24 10 or close to 6.34 · 10 1 3 possible combinations. If each computation takes 1ms to compute [10], 2009 years are needed to evaluate all possibilities. This is only considering the cases with equal rates for each, and only for HSDF graphs.…”

“…Besides, even for one application, the number of ways extra edges can be added to model actor dependency is exponential. For multiple applications the number of computations is huge; and if we need to obtain throughput of the graph, an HSDFG can take too much time to analyze and provide results [10]. Further, only static order arbitration can be modeled using this technique while the best performance of SDFG applications is obtained in a self-timed mechanism [21].…”

“…• Actor array: A list of actors along with their execution times, repetition vector [10] and node to which they are mapped to.…”

Exploiting Segregation in Bus-Based MPSoCs to Improve Scalability of Model-Checking-Based Performance Analysis for SDFAs

“…This in turn may lead to performance problems for the analysis methods. Ghamarian [8] presented novel methods to calculate performance metrics for single SDF applications which avoid translating SDFGs to HSDFGs. Nevertheless, resource sharing and other architecture properties were not considered.…”