SysPad › Examples › VFX render farm
VFX render farm
A submission API queues frame-render jobs onto AWS Batch, which schedules them across an EC2 fleet sharing FSx for Lustre scratch storage during the render.
Job-shaped (not request-shaped) throughput, and Lustre throughput scaling with provisioned capacity rather than a fixed MB/s.
- Load: 2 req/sec
- First to saturate: Render Farm (Batch)
- Estimated cost: $72k/mo
How it works
- Submission API
- Job submissions, not page loads. A couple per second is a busy studio, and the throttle is there to stop one runaway script queueing a million frames.
- Render Farm (Batch)
- Schedules frames onto the fleet and scales it back to nothing between renders. The queue is the point: submissions arrive in bursts and each job runs for minutes, so no artist waits on a scheduler.
- Lustre Scratch
- Shared scratch space every render node reads and writes during a job. Lustre throughput scales with provisioned capacity, which is the whole reason to pay for it over EFS or plain S3 here.
Lessons for this architecture
- Modelling users and traffic entry points
- How Amazon API Gateway works
- How AWS Batch works
- How FSx works
- How S3 works