Evolution of Bilibili Public Network Architecture: From 1.0 to 2.0 and CDN Integration

The Bilibili System Department Network Team is responsible for planning, designing, building, operating, and optimizing the company's data‑center networks, covering internal and external networks, backbone, load balancing, transport, virtualization, and international connectivity.

01 Introduction

According to the Q3 2022 financial report, Bilibili's MAU reached 330 million. To support this growth, the basic network team has continuously optimized the Internet access architecture, upgrading the public network over the past two years to improve stability and cost efficiency.

02 Bilibili Public Network 1.0

The 1.0 architecture features independent static bandwidth lines per IDC, with critical services using BGP bandwidth for low‑latency requirements. The core network devices act as the hub, with load balancers (LB), NAT, and DCN networks attached.

The 1.0 design is simple but has several issues:

Network fault domain: a single core device failure can affect the entire network.

Reliability: heavy dependence on ISP networks; fault isolation is limited.

Scalability: static bandwidth expansion is slow and costly.

Resource constraints: IPv4 scarcity and uneven bandwidth utilization across IDC sites.

Bandwidth cost: growing BGP bandwidth usage raises expenses.

Disaster recovery: regional ISP outages can break same‑city active‑active setups.

03 Bilibili Public Network 2.0

To address the above problems, the team researched leading internet companies and redesigned the network, focusing on regionalization, differentiated SLA, resource centralization, and hardware specialization.

The new 2.0 architecture includes an external backbone and IDC public network, with the following improvements:

Regionalized network zones with clear responsibilities.

SLA‑based design for different service requirements.

Centralized POP nodes that aggregate multiple ISP bandwidths.

Use of routers for the external backbone and switches for internal networks.

Two‑arm LB/NAT deployment for better isolation.

Clustered LB for high availability and shared public IPs.

SGW (Security Gateway) for traffic inspection.

The 2.0 network was fully deployed and stabilized by early 2022.

04 External Backbone Expansion – CDN Network

Bilibili leverages a multi‑level CDN architecture to deliver video content efficiently. The hierarchy consists of edge CDN nodes, regional backbone nodes, and core IDC source sites.

Three CDN back‑haul strategies have evolved:

Early stage: direct public‑network back‑haul (high cost, high risk).

Mid stage: hierarchical CDN with nearby public‑network back‑haul to reduce latency and bandwidth usage.

Later stage: dedicated private lines for back‑haul of selected CDN nodes, minimizing public‑network dependence. The CDN hierarchy (edge → regional → core) provides scalable, low‑latency delivery while controlling costs. 05 Future Outlook The network team will continue to improve stability, cost, and efficiency, expanding external backbone coverage, enhancing cross‑region public‑network scheduling, and exploring advanced technologies to further lower latency and operational expenses. References: [1] B4 and After: Managing Hierarchy, Partitioning, and Asymmetry for Availability and Scale in Google Software‑Defined WAN. [2] Engineering Egress with Edge Fabric: Steering Oceans of Content to the World. [3] Building Express Backbone: Facebook’s new long‑haul network.