- 30 views
- 40 downloads
Model Predictive Congestion Control in RDMA-Enabled High-Speed Datacenter Network
-
- Author / Creator
- Zheng, Yiming
-
With the growing demand for online applications such as high-resolution video streaming and cloud gaming, there is an urgent need for high-throughput and low-latency technologies. The Remote Direct Memory Access (RDMA) over Converged Ethernet Version 2 (RoCEv2) network protocol is increasingly preferred in Data Center Networks (DCNs) as it reduces overhead from processor utilization, an issue with the traditional TCP/IP protocol. Along with these technological advancements, effective congestion control (CC) algorithms are vital. Without proper CC, DCNs could experience performance degradation, affecting the Quality-of-Service (QoS). Traditional CCs are not optimized for DCNs as they were developed for Wide Area Networks (WANs). Even though some CCs, such as TIMELY, were designed specifically for DCNs, they struggle with inevitable packet queuing that causes extra latency. High Precision Congestion Control (HPCC) uses in-flight byte estimation for proactive control and performs better than its peers. However, HPCC, like all other CCs, does not employ a mathematical dynamic equation to evaluate network conditions for optimal control actions. This research dives deep into the fundamental dynamics of DCNs and develops a state-space model to explain network behaviors. Leveraging this model, we present the Model Predictive Congestion Control (MPCC). This new method incorporates constrained mathematical optimization and model predictive control (MPC) design principles. As a window-based CC, MPCC adjusts its window size dynamically based on observed network conditions. In experiments ranging from bursty incast to real-world traffic patterns, we show that MPCC significantly reduces switch queue length and flow completion time while maintaining fairness and throughput.
-
- Subjects / Keywords
-
- Graduation date
- Spring 2024
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.