Bifrost: Analysis and Optimization of Network I/O Tax in Confidential Virtual Machines

Authors: 

Dingji Li, Institute of Parallel and Distributed Systems, SEIEE, Shanghai Jiao Tong University; Engineering Research Center for Domain-specific Operating Systems, Ministry of Education, China; MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University; Zeyu Mi, Chenhui Ji, Yifan Tan, and Binyu Zang, Institute of Parallel and Distributed Systems, SEIEE, Shanghai Jiao Tong University; Engineering Research Center for Domain-specific Operating Systems, Ministry of Education, China; Haibing Guan, Shanghai Key Laboratory of Scalable Computing and Systems, Shanghai Jiao Tong University; Haibo Chen, Institute of Parallel and Distributed Systems, SEIEE, Shanghai Jiao Tong University; Engineering Research Center for Domain-specific Operating Systems, Ministry of Education, China

Abstract: 

Existing confidential VMs (CVMs) experience notable network performance overhead compared to traditional VMs. We present the first thorough performance analysis of various network-intensive applications in CVMs and find that the CVM-IO tax, which mainly comprises the bounce buffer mechanism and the packet processing in CVMs, has a significant impact on network I/O performance. Specifically, the CVM-IO tax squeezes out virtual CPU (vCPU) resources of performance-critical application workloads and may occupy more than 50% of CPU cycles. To minimize the CVM-IO tax, this paper proposes Bifrost, a novel para-virtualized I/O design that 1) eliminates the I/O payload bouncing tax by removing redundant encryption and 2) reduces the packet processing tax via pre-receiver packet reassembly, while still ensuring the same level of security guarantees. We have implemented a Bifrost prototype with only minor modifications to the guest Linux kernel and the userspace network I/O backend. Evaluation results on both AMD and Intel servers demonstrate that Bifrost significantly improves the performance of I/O-intensive applications in CVMs, and even outperforms the traditional VM by up to 21.50%.

USENIX ATC '23 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.

This content is available to:

BibTeX
@inproceedings {288796,
author = {Dingji Li and Zeyu Mi and Chenhui Ji and Yifan Tan and Binyu Zang and Haibing Guan and Haibo Chen},
title = {Bifrost: Analysis and Optimization of Network {I/O} Tax in Confidential Virtual Machines},
booktitle = {2023 USENIX Annual Technical Conference (USENIX ATC 23)},
year = {2023},
isbn = {978-1-939133-35-9},
address = {Boston, MA},
pages = {1--15},
url = {https://www.usenix.org/conference/atc23/presentation/li-dingji},
publisher = {USENIX Association},
month = jul
}

Presentation Video