Placing Streaming Applications with Similarities on Dynamically Partially Reconfigurable Architectures

By means of partial reconfiguration, parts of the hardware can be dynamically exchanged during operation what allows to adapt the system to changing requirements, and even enables the implementation of self-managing systems. This however requires sophisticated system architectures as well as proper algorithmic runtime support. In this paper, we present an algorithm for placing streaming applications at runtime. The approach considers the heterogeneity of common FPGAs, such as Block-RAMs, as well as the routing restrictions of on-chip streaming interconnections. To reduce reconfiguration time, we extend the data flow graphs by OR-nodes to describe differing parts of applications while keeping their similarities. This allows us to model systems which only reconfigure differing parts when switching between applications. The proposed algorithm is implemented as runtime support on an FPGA-based system-on-chip.