5G NR is designed to operate over a broad range of frequency bands and support new applications with ultra-low latency requirements. To support its extremely diverse operating conditions, multiple OFDM… Click to show full abstract
5G NR is designed to operate over a broad range of frequency bands and support new applications with ultra-low latency requirements. To support its extremely diverse operating conditions, multiple OFDM numerologies have been defined in the 5G standards. Under these numerologies, it is necessary to perform scheduling with a time resolution of $\sim 100 \mathrm {\mu s}$ . This requirement poses a new challenge beyond existing LTE and cannot be satisfied by any existing LTE schedulers. In this paper, we present the design of GPF+, which is a GPU-based proportional fair (PF) scheduler with timing performance under $100 \mathrm {\mu s}$ . GPF+ is an improvement over our GPF in Huang et al. (2018). The key ideas include decomposing the original scheduling problem into a large number of small and independent sub-problems and selecting a subset of sub-problems from the most promising search space to fit into a GPU. By implementing GPF+ on an off-the-shelf NVIDIA Tesla V100 GPU, we show that GPF+ is able to achieve near-optimal PF performance with timing performance under $100 \mathrm {\mu s}$ . GPF+ represents the fastest GPU-based PF scheduler that can meet the new real-time requirement in 5G NR.
