Gaming
Run game servers close to players across regions, keep player state in Redis, put DDoS protection in front and swap entry IPs quickly when one is attacked.
The challenge
- 01Player counts spike at launch and during events.
- 02Volumetric attacks are routine in gaming.
- 03Latency defines the experience, and players are spread out.
Reference architecture
How it rolls out
- 01
Deploy close to players
Place game servers where players are and interconnect regions over private links.
- 02
Block attacks first
Put DDoS protection in front, with scrubbing switching in automatically under attack.
- 03
Keep state in cache
Rooms, leaderboards and sessions live in the cache; the database keeps only what must last.
- 04
Ship updates over the CDN
Client updates and asset packs go out over the CDN so launch day does not swamp the origin.
How a customer uses it
A mobile game studio operating across Greater China and South-East Asia
- Before
- In launch week they were hit by two DDoS attacks and players dropped en masse; cross-region parties often saw over a hundred milliseconds of latency.
- After
- With DDoS protection, nearby deployment and private interconnects, later attacks were stopped at the scrubbing layer and cross-region latency fell by more than half.
−55%
Cross-region latency
99.95%
Launch-month availability
0
Outages from attacks
Design notes
Deploy in several regions
Beijing and Ningxia cover players in Mainland China, Hong Kong and Singapore cover Hong Kong, Macau and Southeast Asia, and London and Texas look after Europe and the Americas.
Patches over the CDN
Deliver client updates and asset packs through the CDN so they do not compete with game traffic.
Bare metal for heavy loads
Latency-critical match servers can run on bare metal with dedicated hardware.