Abstract: Dynamic bandwidth allocation (DBA) is a fundamental challenge in the realm of networking. The rapid, accurate, and fair allocation of bandwidth is crucial for network service providers to fulfill service-level agreements (SLAs), alleviate link congestion, and devise strategies to counter network attacks. However, existing bandwidth allocation algorithms mainly operate on the control plane of the software-defined networking (SDN) paradigm, which can lead to considerable probing overhead and convergence latency. Moreover, contemporary network architectures necessitate a hierarchical bandwidth allocation system that addresses latency requirements. We introduce a finegrained, hierarchical and scalable DBA algorithm – the HSDBA algorithm – implemented on the programmable data plane. This algorithm reduces network overhead and latency between the data plane and the controller, and it is proficient in dynamically adding and removing network configurations. We investigated the practicality of HSDBA using protocol oblivious forwarding (POF) software switches. Experimental results show that HSDBA achieves fair bandwidth allocation and isolation guarantee within approximately 25 packets. It boasts a convergence speed 1.4 times faster than the most recent algorithm, namely, AHAB; meanwhile, it maintains a bandwidth enforcement accuracy of 98.1%.