How ByteDance’s STE Team Tackles Linux Kernel Memory Waste and Drives Cloud‑Native Innovation
This interview reveals how ByteDance’s STE team, a youthful group of Linux kernel engineers, identified and solved memory‑management redundancy, contributed the HVO and VDUSE projects to the open‑source community, and leveraged these advances to boost cloud‑native performance and reliability across the company.
ByteDance’s STE (System Technologies & Engineering) team, formed in 2015 and officially named in 2018, grew from a handful of engineers to a multi‑hundred‑person group with R&D centers worldwide, focusing on Linux kernel, firmware, compiler technology, virtualization, networking, and intelligent operations.
Young Team Behind Linux Kernel
The core kernel team, largely composed of post‑90s engineers, provides Linux kernel services for all internal ByteDance businesses, handling kernel management, process scheduling, virtualization, and networking. They emphasize nurturing young talent through mentorship and opportunities.
Solving Decades‑Old Memory Management Redundancy
Linux manages physical memory in 4 KB pages, even when using large pages (2 MB or 1 GB), leading to wasted struct page allocations. STE engineer Song Muchun led the HVO (HugeTLB Vmemmap Optimization) project to reduce this redundancy, potentially saving up to 16 GB per terabyte of server memory.
The year‑long effort (2020‑2021) involved code research, development, testing, and refactoring, ultimately gaining industry recognition from Huawei, Google, AWS, and Oracle.
Cloud‑Native Device Virtualization with VDUSE
To bridge the gap between container I/O limitations and traditional virtualization, STE created the VDUSE framework, enabling user‑space creation of virtual devices that integrate with the vDPA, virtio, and vhost subsystems. VDUSE was open‑sourced in 2020 and merged into Linux 5.15.
VDUSE devices are created via ioctl calls on /dev/vduse/control, exposing character devices for configuration and data exchange, and are now deployed at scale in ByteDance’s cloud‑native environment.
Community Engagement and Future Vision
The team stresses open‑source collaboration, iterating HVO through more than 20 versions based on community feedback, and aims to explore further cloud‑native and high‑performance networking scenarios. Their long‑term goal is to push foundational infrastructure forward while fostering a culture of mentorship and innovation.
ByteDance SYS Tech
Focused on system technology, sharing cutting‑edge developments, innovation and practice, and analysis of industry tech hotspots.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.