Boutique studios often grow a render farm one machine at a time until it becomes hard to manage. A few decisions made early keep it fast, reliable and easy to scale.
Manage nodes as a fleet
Build render nodes from a standard image and manage them with configuration management tools such as Ansible or Puppet. Every node then has the same software versions, plugins and settings, and a failed node can be rebuilt in minutes instead of hours.
Storage is usually the bottleneck
When many nodes read the same textures and caches at once, shared storage slows everything down. Size storage for the total throughput of the farm, not just its capacity, and keep caches and scratch data on fast local disks where possible.
Network for throughput
Gigabit networking is quickly outgrown. Moving storage and busy nodes to 10 or 25 GbE, with a switch that can carry the load, often gives a bigger speed-up than new processors.
Schedule and monitor
- Use a render manager such as Deadline to queue, prioritise and retry jobs.
- Monitor node health, disk space and licence use so failures are caught overnight, not in the morning.
- Keep licence servers highly available, since a licence outage stops the whole farm.
Think about security early
If clients will ask for a content-security assessment, design the network segmentation and access controls from the start. Retrofitting them is much harder.
Linux infrastructure is our home ground. Call 04 282 0045 or see how we help studios.