Live streaming and video
Ingest, transcoding, delivery and storage end to end: GPU-accelerated transcoding, CDN delivery close to viewers, sources and replays in object storage, scaled for peaks.
The challenge
- 01Streams stutter and lag at peaks, and viewers leave.
- 02Transcoding eats compute — idle most of the time, short at peaks.
- 03Replays and assets pile up and storage costs run away.
Reference architecture
How it rolls out
- 01
Ingest close to the source
Streamers push to the nearest region to cut first-mile latency.
- 02
GPU transcoding
Multi-bitrate transcoding on GPUs, with daily GPUs added before peaks.
- 03
Deliver over the CDN
Viewers pull from the nearest edge; buy traffic packages for predictable volume.
- 04
Replays and archive
Replays in object storage, older material moved to infrequent access automatically.
How a customer uses it
A cross-border live-commerce agency
- Before
- Big streams stuttered for viewers, monthly transcoding servers sat idle, and replays filled the disks.
- After
- Daily GPUs for transcoding, CDN delivery and replays in object storage: stutter complaints fell sharply and transcoding costs only arise on streaming days.
−85%
Stutter complaints
−50%
Transcoding cost
Automatic
Hot-to-cold tiering
Design notes
Daily GPUs for transcoding
Add daily GPUs for transcoding before big events and delete them afterwards.
Hot and cold tiers
Recent replays in standard storage, older material moved to infrequent-access storage automatically.
Traffic packages save money
Buy traffic packages for predictable delivery; it costs less than pay-as-you-go.