.png){kind=icon}
.png){kind=icon}
Return
A Fast-Converging Low-Latency RDMA Congestion Control Algorithm
.png){kind=icon}
PRE
AI
DOI：10.1109/MNET.2025.3589913.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Data center networks are critical infrastructures currently serving fields such as cloud computing and artificial intelligence, where congestion control algorithms play a critical role in improving network performance. In this paper, we focus on the current research status and prospects of congestion control approach in Remote Direct Memory Access (RDMA) networks and propose a new algorithm to address existing shortcomings. Firstly, we classify and summarize current congestion control algorithms, exploring the advantages and disadvantages of the end-to-end and end-network collaboration approaches. We argue that, considering the trade-off between network performance and hardware costs, attention should still be directed towards end-to-end congestion control algorithms. Then, we analyzed issues with mainstream solutions and introduced a new fast convergence congestion control algorithm, FCC. This algorithm is based on Explicit Congestion Notification (ECN) and Round-Trip Time (RTT) signals, leveraging the gradient of RTT and a Sigmoid curve to enhance response speeds. Finally, we conducted a performance evaluation of this algorithm and compared it with existing ones. Experimental results demonstrate that the FCC algorithm achieves superior performance in terms of convergence speed, fairness, and latency for short flows.
Keywords:
RDMA congestion control
RTT
ECN
Sigmoid-AI
Journal
IF:
6.3
Papers:
2.6K
Citations:
1.1W