Resource Positioning and Sharing in Dynamically Reconfigurable FPGA-Based Networks

Using FPGAs as processing units to run Virtualized Network Functions (VNFs) in Service Function Chains (SFCs) is a promising way to achieve flexibility aligned with the high performance demanded by many VNFs. However, their efficient use requires defining where to place such devices in a network and how their resources are shared among multiple VNFs. Moreover, as demand for different VNFs may vary over time, device-level adaptability may be required to better accommodate current demands. In this work, we present an ILP-based approach able to maximize the total value of allocated SFCs through optimized FPGA positioning and partitioning within a network. Moreover, we demonstrate how device repartitioning can be used to accommodate varying user demands over time. We demonstrate that, on average, a 15.3% improvement in total value can be achieved, compared to static partitioning.