Towards Models for Many-Cores: The Case for the Reconfigurable Mesh

The reconfigurable mesh has been extensively studied as a theoretical model for massively parallel computing, but its implementation has been severely hindered by the unrealistic assumption of constant time broadcasts. Nevertheless, we investigate the reconfigurable mesh as a model for future many-cores due to its many interesting features. In this paper, we discuss the realization of such a many-core architecture on FPGA technology and a corresponding tool flow including a new programming language. Then we focus on two important characteristics for a many-core, scalability and fault tolerance. We present reconfigurable mesh algorithms to achieve these characteristics and elaborate on the new challenges faced when applying these algorithms to practical reconfigurable meshes.