01
Web serving
Request handling, caching and application runtime tuned for continuous high volume rather than for a burst.
Infrastructure
The layer everything else stands on.
Web serving, delivery, storage and compute, engineered and operated directly rather than assembled from managed services whose behaviour we would not control.
- Serving
- Tuned for sustained request volume
- Delivery
- Multi-provider, own origins
- Compute
- Including GPU, specified in house
- Access
- Key-based authentication only
01 / Components
What we operate.
02
Databases
Transactional and analytical stores, each chosen for its workload rather than one store asked to do both.
03
Content delivery
Multiple delivery providers in front of origins we control, with routing and cache policy managed per region.
04
Storage
Object and block storage sized for large media libraries, with lifecycle and retention applied deliberately.
05
GPU compute
Accelerated compute for inference and generation workloads, specified, scheduled and maintained in house.
06
Operations
Monitoring, alerting, change logging and recovery procedures, treated as part of the product rather than as an afterthought.
02 / Operating practice
Changes are recorded, backups are proven.
Configuration and infrastructure changes are written to an internal change log as they happen, so the state of a system is a record rather than a recollection.
A backup that has never been restored is not a backup. Backups are verified by replaying them into a scratch environment and counting what comes back, not by checking that a file exists.
Access to production is by key-based authentication only, and privileged actions are logged.
O1
Change log
Every configuration and infrastructure change recorded at the time it is made.
O2
Verified backups
Restored into a scratch environment and checked, on a schedule.
O3
Key-based access
No password authentication to production systems.
O4
Documented recovery
A written restoration procedure that has been followed, not just written.