CONCEPT Cited by 1 source
Multi-tenant capacity management¶
Multi-tenant capacity management is the discipline of allocating, isolating, and sharing compute resources across multiple organizational tenants with guarantees about minimum entitlements and fair access to surplus.
Definition¶
In a shared compute platform serving diverse teams, capacity management must balance three tensions: (1) isolation — tenants should not starve each other; (2) utilization — idle reserved capacity should not go wasted; (3) priority — business-critical workloads should run first. Solutions typically combine reserved quotas, shared pools, fair-sharing weights, and preemption policies.
At Netflix¶
Netflix manages capacity through a hierarchical tenant model with two capacity types:
- Reserved — partitioned exclusively for a tenant (leaf) or fair-shared across a subtree (internal).
- Shared — a global pool any tenant can burst into beyond its reservation.
With Kueue, idle reserved capacity is lent to other tenants and reclaimed via preemption when the owner needs it back — a semantic improvement over the original CMB system where admission-only fair sharing left idle reservations unused.
(Source: sources/2026-06-22-netflix-how-netflix-simplified-batch-compute-with-kueue)
Seen in¶
- sources/2026-06-22-netflix-how-netflix-simplified-batch-compute-with-kueue — Netflix's CMB/Kueue capacity model with reserved + shared pools and preemption-based reclamation.