How High‑Performance IaaS Virtualization Powered Douyin’s Spring Festival Red‑Envelope Event

ByteDance cloud engineer Wu Chen discusses the role of system‑level IaaS virtualization in supporting the massive resource demands of Douyin’s Spring Festival red‑envelope activity.

Background

Wu Chen, a mathematics graduate from Shandong University and a computer science master’s from Zhejiang University, joined ByteDance in June 2020, contributing to Singapore data‑center construction and the Douyin Spring Festival event.

Advantages of Virtualization

He highlights that IaaS virtualization offers superior elasticity, cost‑effectiveness, and resource utilization, with strong isolation at the scheduling, memory, and I/O levels, preventing hidden resource contention and masking hardware differences. System‑level virtualization abstracts CPU, memory, and I/O into independent virtual machines with their own OS and IP.

In contrast, containers are OS‑independent environments that isolate applications using Linux namespaces and cgroups, making them lighter weight but offering less isolation than virtual machines.

Challenges During the Red‑Envelope Event

The event generated tens of millions of QPS, requiring rapid scaling and resource sharing. Virtualization enabled elastic scaling, rapid VM switching for time‑sharing, and hot migration to minimize downtime during failures.

A power‑outage test at the Shenyang aggregation data‑center revealed unexpected rack power loss, serving as a real‑world reliability exam that confirmed the system’s ability to detect, respond to, and recover from faults just before the event.

Future Outlook

With emerging large‑scale hardware from AMD and Intel, virtualization will continue to improve resource utilization, simplify scheduling, avoid resource competition, and boost performance, but its evolution will remain tightly coupled with hardware advancements.

For a deeper dive into high‑performance IaaS virtualization practices, Wu Chen invites readers to attend his upcoming meetup where he will share specific optimizations that reduce virtualization overhead to near‑bare‑metal performance.