Haonan Li, Tian Pan, Jin Ke, Baohai Hu, Changgang Zheng, Enge Song, Zhi Xu, Ye Yang, Bowen Yang, Donglin Lai, Yisong Qiao, Bengbeng Xue, Jianyuan Lu, Xiaoqing Sun, Shize Zhang, Zihao Fan, Mingxin Li, Yang Song, Jun Liang, Xionglie Wei, and Biao Lyu, Alibaba Cloud; Rong Wen, Fudan University and Alibaba Cloud; Zhigang Zong, Alibaba Cloud; Jiao Zhang and Tao Huang, Purple Mountain Laboratories; Shunmin Zhu, Alibaba Cloud
Major cloud providers often build NFV products by reusing existing architectures. These designs were originally developed and optimized for tenant virtual machines (VMs) that operate at network endpoints. In contrast, network function (NF) VMs function as intermediate forwarding nodes shared by multiple tenants, which have very different resource demands. Directly applying tenant-oriented designs creates mismatches between infrastructure capabilities and NF requirements. A common case arises when NF VMs serve many tenants: they typically exhaust vNIC and sessions on vSwitches well before CPU resources are saturated. At this point, NF VMs cannot accept additional traffic despite having idle CPU cycles. Providers usually address this by scaling out or scaling up NF VMs, which wastes resources and increases cost. This is just one example, where numerous other mismatches remain. To address these, we present CStar Gateway. CStar Gateway shifts NF VM multi-tenancy support from vSwitches into NF VMs, which reduces vNIC and session bottlenecks with minimal changes to the existing cloud infrastructure. CStar Gateway also identifies and redirects I/O- or CPU-intensive flows to FPGA-based NFs, increasing the service capacity. In addition, CStar Gateway takes over NF VM elasticity support from vSwitch, simplifying and accelerating the scaling process, and thereby enhancing overall system flexibility. Deployment results show that the design improves CPU and I/O utilization of NF VMs by at least 5x and reduces NF cluster capital expenditure by 71.91% to 88.57%.
NSDI '26 Open Access Sponsored by
King Abdullah University of Science and Technology (KAUST)
Open Access Media
USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

author = {Haonan Li and Tian Pan and Jin Ke and Baohai Hu and Changgang Zheng and Enge Song and Zhi Xu and Ye Yang and Bowen Yang and Donglin Lai and Yisong Qiao and Bengbeng Xue and Jianyuan Lu and Xiaoqing Sun and Shize Zhang and Zihao Fan and Mingxin Li and Yang Song and Jun Liang and Xionglie Wei and Biao Lyu and Rong Wen and Zhigang Zong and Jiao Zhang and Tao Huang and Shunmin Zhu},
title = {{CStar} Gateway: Augmenting Public Cloud Infrastructure for Heterogeneous Network Function Virtualization},
booktitle = {23rd USENIX Symposium on Networked Systems Design and Implementation (NSDI 26)},
year = {2026},
isbn = {978-1-939133-54-0},
address = {Renton, WA},
pages = {1501--1515},
url = {https://www.usenix.org/conference/nsdi26/presentation/li-haonan},
publisher = {USENIX Association},
month = may
}