Technical Assurance Practices for the 13th League of Legends World Championship Live Stream

The article shares the technical assurance practice and thinking for the 13th League of Legends World Championship (S13) live stream on Bilibili, whose business goal is to reach 120 million viewers.

Three stages are defined: pre‑event traffic accumulation, in‑event user experience in the main room, and post‑event replay and interaction. The main challenges are peak‑traffic estimation and the wide scope of business functions that must be protected.

The assurance plan is built by first mapping all business functions (over 60) and their core metrics (e.g., PCU – peak concurrent users). These functions are grouped into Activity Page, Traffic Entry, Main Room, and Replay Page, each with its own traffic characteristics.

A traffic‑estimation model converts business metrics (exposure, conversion rates) into technical metrics (QPS/TPS). For example, QPS = exposure × conversion × … / duration, which yields the expected request rate for each scenario.

Typical scenarios are described in detail:

Room‑entry: users arrive from splash screens, recommendations, push notifications, etc.; QPS is the sum of entry‑point QPS.

"Selected Moment" feature: a pop‑up that lets users participate; QPS = PCU × participation‑click rate, mitigated by traffic‑shaping.

Long‑connection: chat, gifts, and other real‑time interactions; pressure = N × PCU, controlled by isolating long‑connection traffic.

Exit scenarios: returning to entry pages or sliding to another live room; both generate spike traffic that is throttled or cached.

Global traffic is monitored both at the business‑scenario level and the whole‑event level, allowing capacity planning and resource pre‑allocation.

Task assignment follows the RASIC principle: each protection item has a responsible owner, a department‑level coordinator, and a regular sync mechanism.

Key practice areas include:

Technical chain mapping – using the Advisor platform to capture user flows, auto‑generate dependency graphs, and calculate amplified QPS/TPS.

Fault‑injection drills – using the Fault platform to identify strong and weak dependencies, inject failures, and verify that core functions remain unaffected.

Full‑link load testing – with the Melloi platform, focusing on hotspot keys, empty‑cache protection, and consumption back‑pressure.

Pre‑incident SOP – defining 1‑minute detection, 5‑minute定位, and 10‑minute recovery steps.

Change control – strict change‑gate during the event, with a green‑channel for emergency changes.

In‑event tracking – real‑time dashboards for SLO, PCU, QPS, latency, and resource water‑marks.

The outlook suggests turning the accumulated experience into a platform‑level solution for future large‑scale live events.

In conclusion, S13 achieved over 120 million viewers and a peak of 460 million concurrent users, meeting its business goals thanks to the systematic technical assurance described above.