Connecting CephFS and Parallel Filesystems to Pods

Large-scale scientific instruments like ALS beamlines generate petabytes of raw data that must be accessible by containerized analysis pipelines without bottlenecking I/O.

Available Storage Classes

Berkelium provides three storage classes tailored to different workload needs:

1. cephfs-replicated (Default): Shared multi-writer storage (ReadWriteMany) suitable for cross-pod dataset sharing, configuration directories, and notebook home folders.
2. nvme-scratch-local: Ultra-high IOPS local solid-state drives (ReadWriteOnce) mounted on compute nodes, optimal for temporary checkpointing and shuffle files.
3. lbnl-project-archive: High-density bulk cold storage for long-term experiment logs and dataset dumps.

Sample PVC Definition

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: als-beamline-data-pvc
  namespace: sci-als-analysis
spec:
  accessModes:
    - ReadWriteMany
  storageClassName: cephfs-replicated
  resources:
    requests:
      storage: 500Gi

Mount this claim in your pod under /data to begin streaming detector frames directly into your analysis container.