What Chamber Does
Capacity Management
Allocate GPU resources through hierarchical teams with guaranteed reservations
Intelligent Scheduling
Two-phase scheduling with reserved and elastic workload classes for optimal resource utilization
Real-time Visibility
Dashboard with GPU utilization metrics, workload status, and team performance insights
Multi-Cluster Support
Manage capacity across multiple Kubernetes clusters from a single control plane
How It Works
Chamber operates as a SaaS control plane with a lightweight agent deployed in your Kubernetes clusters:1
Deploy the Agent
Install the Chamber agent in your GPU clusters via Helm chart. The agent syncs workload state with the control plane.
2
Model Your Organization
Create teams in a hierarchy that reflects your organization structure.
3
Allocate Capacity
Assign GPU reservations from your capacity pools to teams. Teams get guaranteed access to their reserved capacity.
4
Submit Workloads
Workloads submitted to your cluster are scheduled by Chamber. Reserved workloads get guaranteed resources; elastic workloads use idle capacity.
Key Concepts
| Concept | Description |
|---|---|
| Team | Hierarchical unit representing a team or project. Teams can have projects. (aka child teams ) |
| Capacity Pool | A GPU resource pool, typically a Kubernetes cluster with GPU nodes. |
| Reservation | Links a team to a capacity pool with guaranteed GPU allocation. |
| Reserved Workload | Workload that runs on reserved capacity. Cannot be preempted. |
| Elastic Workload | Workload that uses idle capacity. Can be preempted when reserved workloads need resources. |