A Cluster-based Hierarchical Partitioning Approach for Multiple FPGAs

摘要

—Most high performance computing systems arelarge-scale computing systems, and consist tens ofthousands computing nodes with superior capabilities.FPGAs are able to accelerate large scope and complicatedcomputing with flexible configurations. More and morecompanies and research institutions integrate multi-FPGAsinto high performance computing systems to get a bettertrade-off between high-performance and low power. How todesign an effective topology for these integratedmulti-FPGAs according different applications has become akey problem in this area. Acluster based architectureandcorresponding partitioning approach are proposed in thispaper. The proposed hierarchical topology taking fulladvantages of both traditional metallic lines and emerginginterconnections to implement one-hop local communicationwithin the cluster and one-hop global high-speedcommunication between clusters. The case study provedthat the proposed architecture and partitioning approachcan implement the fast mapping from the design to realcomputing system with multi-FPGAs, and accelerate therealization of high performance reconfigurable computingsystems.