ABSTRACT Modern model‐based congestion control protocols such as BBR achieve high throughput by accurately estimating bottleneck bandwidth and RTT. However, during ProbeBW phases, bandwidth ramp‐up can introduce transient queues that accumulate over successive cycles, leading to tail latency increase and degraded performance for delay‐sensitive applications like video streaming. In this paper, we propose active queue debt compensation (AQDC‐BBR), a lightweight extension to BBR that explicitly quantifies and models the queue debt incurred during bandwidth probing and compensates it in subsequent phases. We evaluated AQDC‐BBR in both NS‐3 simulations and Mininet video streaming experiments. The results show that AQDC‐BBR reduces tail latency by 26% and 25% compared to Copa and PCC, respectively, and achieves a reduction of more than 25% compared to BBR and its variants, while maintaining stable high throughput.

Remote Direct Memory Access (RDMA) is becoming a popular high-speed networking technology. It uses kernel bypass and zero copy to achieve high throughput and low latency with little CPU overhead. However, standard RoCE transmission uses Equal Cost Multipath (ECMP) for load balancing, which can result in lower transmission performance due to hash conflicts. Meanwhile, it has been verified that, unlike TCP, the unique retransmission mode and flow characteristics of RoCE make previous load balancing algorithms not well applied to RoCE. In this paper, we introduce RoCELet, a load balancing algorithm for RoCE. It achieves fine-grained RoCE load balancing by actively generating flowlets, effectively utilizing the rich end-to-end paths in the data center. We implement a prototype based on DPDK and evaluate it through small-scale testbed experiments and large-scale simulations. Our results show that compared to state-of-the-art load balancing algorithms, RoCELet optimizes 48.2% and 16.4% in average FCT and$99^{th}$-ile FCT, respectively.