How Mango TV Built a Hybrid Cloud Platform to Scale Video Services

Compared to Internet Companies, Traditional Media Development Prioritizes Stability

Peng ZhefU, formerly of Douban where he led DoubanAppEngine, joined Mango TV and observed the contrast between fast‑paced internet product development and the more cautious, stability‑focused approach of state‑owned media platforms.

Mango TV’s core business spans three lines: local IPTV, OTT with millions of paying users, and website content that streams all Hunan TV programs online. The platform also pioneers live‑stream innovations such as global Billboard broadcasts and multi‑camera New Year’s Eve shows.

Evolution of Mango TV’s Platform Architecture

With ample resources, Mango TV has been able to adopt advanced technologies ahead of many internet companies. The team migrated from SVN to Git, introduced large‑scale Redis‑Cluster deployments, and embraced Docker to simplify infrastructure.

They built a Docker‑based scheduling platform and, without referencing existing orchestration systems, created their own scheduler that drives Redis clusters, achieving millisecond‑level scaling and nine‑nines availability.

1. Building a Proprietary Platform to Support Core Business

Leveraging his Douban experience, Peng created a new PaaS called NebuliumEngine (NBE), featuring a two‑level Nginx architecture, service discovery via SkyDNS, and configuration stored in etcd. Runtime isolation is achieved entirely with Docker, and control is moved outside the containers.

Challenges arose because the organization lacked a unified strong language, leaving runtime control to business teams and causing resource‑management difficulties.

The first‑generation NBE architecture is illustrated below:

At the end of 2014, the team revisited Borg and Omega concepts and began developing the second‑generation NBE, now known as ProjectEru, a Borg‑like service orchestration and scheduling platform.

ProjectEru can mix offline and online services, allocate CPU resources with fine‑grained granularity (0.1, 0.01, 0.001), and uses Redis as a data bus for message publishing and container state monitoring. It integrates Docker ImageLayer features with Git versioning for automated build/test pipelines, unifies deployment environments, and eliminates runtime pollution, enabling rapid scaling.

The architecture of ProjectEru is shown below:

Unlike Kubernetes, ProjectEru’s Pods serve only logical isolation for business grouping, while actual isolation is enforced at the network layer. Business teams cannot write Dockerfiles directly; instead, an App.yaml standardizes Dockerfile generation, and a common Entrypoint enables code reuse across roles.

ProjectEru abandons the closed‑loop design of the first‑generation NBE, breaking the monolithic Master (Dot) into stateless cores that can scale out independently, improving reliability.

2. Introducing Public‑Cloud Services to Build a 360‑Degree Solution

To balance core and edge workloads without letting spikes affect core services, Mango TV adopted a hybrid cloud model: a private cloud for core operations and public‑cloud services for bursty demand. The private platform could no longer meet all scaling needs, and building additional data centers would waste resources.

Qiniu was selected as the public‑cloud provider, offering a 360‑degree solution for both streaming and live‑broadcast scenarios. Their optimized EC technology reduces storage redundancy from three copies to 1.125, achieving 16‑nines reliability. Data is split across servers to ensure resilience, and media files are stored as encrypted slices.

Qiniu also provides fast transcoding, watermarking, video clipping, and format conversion, along with a pooled compute resource that can dynamically expand or shrink. Their self‑developed multi‑CDN management platform gives users transparent, self‑service monitoring and health management across upload, storage, download, and distribution stages.

Conclusion

A robust technical platform underpins Mango TV’s rapid growth, and the hybrid‑cloud strategy has saved significant time and cost while enabling swift business expansion within just a year and a half. This case demonstrates how traditional broadcast institutions can leverage cloud services to achieve agile transformation and maintain a competitive edge.