Methodologies for Synthesizing and Analyzing Dynamic Dataflow Programs in Heterogeneous Systems for Edge Computing

The possibility of using the increasing computing power available in cloud infrastructures requires development new approaches for application software and optimization. Emerging edge paradigms offer reducing bandwidth needs optimizing latency, features particularly relevant Big Data applications, by bringing computation closer to user data generation processes. However, pose several challenges terms how be able efficiently take advantage a distributed network heterogeneous processing nodes. This paper deals with this problem extending dynamic dataflow framework related design flow tools support platforms. describes methodology steps synthesis executing on CPU/GPU co-processing do include optimization communication between elements, technique efficient mapping parallelization independent GPU partitions, introduction programming approach leveraging SIMD nature computing. To complete seamless porting partition CPU or nodes, an automated exploring configuration space identify high performance working points is developed.