How PCDN Cuts Bandwidth Costs While Delivering HD Video at Scale

Data shows that domestic internet traffic consumes about 200 EB per month, with 80% coming from video. As 5G spreads, cloud‑based broadcasting grows, driving bandwidth costs higher.

1. Technical Challenges in Reducing Bandwidth Costs

Multiple challenges arise: diverse terminal types (Android, iOS, PC, Web, OTT), varied video services (live, VOD, cache download, short video) each with different performance metrics, numerous usage scenarios (smart caching, edge‑play, speed‑up), and a long‑tail of less‑popular videos that strain storage.

2. Strategy: Cloud‑Edge Content Distribution Network (PCDN)

PCDN is a three‑level network built on P2P technology that aggregates idle fragmented resources to form a high‑quality, low‑cost distribution system.

Three‑Tier Architecture

Level 1: Cloud CDN – high upload capacity, stable service, but expensive bandwidth.

Level 2: Edge network (edge nodes, routers, commercial Wi‑Fi) – moderate capacity, 24 h online, can replace level 1 in many cases.

Level 3: End devices – many, low upload/storage, but can provide nearby sources for popular fragments.

When a device plays video, PCDN selects the optimal tier to download fragments, minimizing cost while ensuring smooth playback.

3. P2P Fundamentals

Nodes exchange video fragments directly. A fragment’s identifier is generated from URL features to create a compact, unique ID. Resources are chunked into fixed‑size pieces, and only hot fragments are cached on edge nodes.

4. Resource Storage

How to identify a resource?

Normalize resources into fixed‑size chunks.

Decide which fragments to store based on popularity and cache size.

Edge nodes receive popularity statistics from the server and store only the most requested fragments, improving cache utilization and P2P sharing rates.

5. Node Allocation

Allocation consists of node filtering, scheduling, and intelligent distribution.

Node Filtering

Filter by NAT type to remove poorly connected nodes.

Filter by node quality to exclude low‑performance peers.

Node Scheduling

Proximity principle: prioritize nodes in the same community, city, province, region, or nation.

Capacity matching: allocate tasks proportionally to each node’s upload capability.

Intelligent Distribution

Dynamic allocation considers client‑side information such as video resolution and buffer level, assigning more high‑quality nodes for high‑resolution streams and more low‑cost nodes when buffer is sufficient.

6. Download Scheduling

The buffer is divided into three zones: urgent (download from CDN), transition (mix of level 2/3), and safe (prefer level 2/3). Strategies adapt to real‑time network conditions, playback point, and historical stall data.

7. Node Management & Task Assignment

Nodes are pre‑fetched for upcoming fragments, scored by first‑packet latency, download speed, and task success, allowing the system to converge on high‑quality peers while discarding poor ones. Tasks follow a “more work, more reward” policy, with early timeout detection and reassignment.

8. Data Sharing

Sharing establishes a connection (direct, reverse, or hole‑punching) between peers. After connection, data is transferred via a custom reliable UDP protocol with congestion control, fast start, loss prediction, and fast retransmission.

Data Verification

Both MD5 (high security) and CRC (low overhead) are combined: critical data uses MD5, non‑critical data uses CRC, ensuring integrity without excessive performance cost.

9. Live Streaming Specifics

Live streams face low latency, tiny buffer, and high dynamics, making P2P sharing harder. Edge nodes act as stable suppliers, synchronizing with CDN to supplement first‑level resources and improve overall sharing efficiency.

10. Practical Experience

Complex network systems should be decomposed into subsystems, each with clear inputs, outputs, and metrics. Modeling (e.g., water‑tank for download scheduling, supply‑demand for node allocation) and rapid iteration—using load tests, sample training, and bucket validation—help optimize parameters and achieve stable performance.