An Interconnect-Centric Approach to the Flexible Partitioning and Isolation of Many-Core Accelerators for Fog Computing

Fog computing platforms are currently challenged by new design requirements: low-overhead sharing mechanisms among Internet-of-Things (IoT) services, the strong isolation of their execution environments and adaptive resource allocation to them, matching the inherent dynamic nature of most IoT applications. This paper targets the fulfillment of these requirements in a many-core programmable accelerator for future Fog computing nodes by leveraging the pivotal role played by the on-chip communication architecture for efficient, flexible and secure sharing of the whole device.