Tencent Classroom Video Cloud Migration: Overview, Implementation, and Continuous Optimization

By the end of 2019, Tencent Classroom completed the migration of all live streaming and on‑demand video services to the cloud. The migration involved roughly 4 million videos totaling about 1,500 TB. The article summarizes the whole migration process and compares key quality metrics before and after the migration.

1. H5 Live and Playback Migration – Initially, Tencent Classroom used QQ group live (UDP‑based low‑latency streaming) without H5 live or VOD support. An intermediate platform was built to integrate H5 live and playback, leveraging the cloud interactive live service which already supported H5 live and recording. Multi‑camera streams were mixed in the cloud to select the appropriate video for playback. The before‑and‑after transformation is illustrated in the comparison table.

2. Migration of Existing Video Assets – The migration faced several challenges:

Massive data scale: ~4 million videos, ~1,500 TB.

Long transcoding cycles: video transcoding is CPU‑intensive; with a cloud capacity of 6 million minutes per day, the process still required several months, especially with added encryption transcoding.

Cross‑platform compatibility: existing players could not fully handle encrypted VOD, and mobile app versions needed careful rollout.

Ensuring a seamless user experience during migration.

To address these, the team adopted a gray‑release strategy, incremental migration, fallback playback (cloud VOD first, then original platform if needed), and rigorous data reconciliation.

3. Video Upload Success Rate Improvement – The original system based on FTN was no longer maintained, resulting in a 92 % upload success rate. After switching to Tencent Cloud’s COS‑based VOD upload, the success rate rose to 99 % thanks to nationwide edge access, content acceleration, high reliability with multi‑zone replication, and dedicated DevOps support.

4. Encrypted Playback – To protect digital rights, the team adopted HLS encryption, which follows the HLS standard, offers good compatibility, and incurs low decryption overhead.

5. Continuous Optimization – After migration, the focus shifted to stability and middle‑platform construction. Redundant backup of live streams was introduced to improve reliability, and per‑room and per‑user stream durations were recorded for monitoring and operations.

In summary, the cloud‑based interactive live service filled the gaps of H5 live capabilities, eliminating the need for separate H5 live transcoding development and operations. WebRTC‑based live streaming on the web reduced latency compared with traditional FLV/HLS, enabling richer two‑way interaction. All live‑playback videos are now generated in the cloud, with VOD providing powerful transcoding and distribution; playback generation time dropped from over an hour to a stable 25 minutes, ending the era of split audio‑video platforms for classroom live and on‑demand services.