Resource Hierarchy & Quotas
Berkelium implements multi-tenant isolation through a structured hierarchical resource model:
graph TD Lab[LBNL Science IT Cluster] --> Div1[ALS Division] Lab --> Div2[Biosciences Division] Lab --> Div3[Computing Sciences / ESnet] Lab --> Div4[Physics & Applied Math] Div1 --> NS1[sci-als-diffraction] Div1 --> NS2[sci-als-microscopy] Div2 --> NS3[sci-ml-structbio] Div3 --> NS4[sci-esnet-telemetry]
Namespace Isolation
Section titled “Namespace Isolation”Every namespace on Berkelium is isolated using:
- Kubernetes
ResourceQuota: Enforces strict ceilings on total CPU cores, memory (GiB), GPU count, and PVC storage volume. LimitRange: Applies default and maximum request/limit values for individual pods.- Calico
NetworkPolicy: Prevents inter-namespace traffic except for explicitly whitelisted ingress endpoints. - RBAC RoleBindings: Restricts team members to their assigned scientific namespaces.
Resource Tiers
Section titled “Resource Tiers”| Tier Name | Max CPU | Max RAM | GPUs | Persistent Storage | Best Suited For |
|---|---|---|---|---|---|
| Development | 4 Cores | 16 GB | 0 | 50 GB | Script prototyping, test runs, CI pipelines |
| Standard Compute | 16 Cores | 64 GB | 0 | 250 GB | Data ETL, simulation sweeps, MPI pipelines |
| GPU Research | 32 Cores | 128 GB | 2x H100 / A100 | 1 TB | PyTorch / JAX deep learning, cryo-EM |
| High Allocation | Custom (128c+) | Custom (1.5TB+) | Up to 8x GPUs | 10 TB+ | Foundation model pretraining & beamline pipelines |
