Storage Concepts

MASS organizes storage in a two-level hierarchy:

VolumeGroup  (logical pool, e.g. 10 TiB total)
├── Volume A (e.g. 1 TiB, POSIX)
└── Volume B (e.g. 500 GiB, HDF5)

VolumeGroup

A VolumeGroup is a logical storage pool provisioned from one or more storage nodes. It defines total capacity, performance profile, and data protection settings.

Profiles

Profile	Use case
Standard	General-purpose workloads
Performance	Low-latency, metadata-intensive workloads (ML/AI training)
Archive	Cost-optimized, high-capacity sequential workloads

Data Protection

When creating a VolumeGroup, select one of three protection modes:

• Single Copy: No redundancy. Best for scratch or temporary workloads where data loss is acceptable.
• Replicated: Keeps multiple copies of all data across nodes. Simple to configure, higher storage overhead.
• Erasure Coded: Splits data into shards with configurable parity (e.g., 8+2 = 2-fault tolerance at 20% overhead). Most storage-efficient for large datasets.

DeletionPolicy

• Retain (default): Deleting the VolumeGroup preserves the underlying data. Re-creating a VolumeGroup with the same label reconnects to it.
• Delete: Deleting the VolumeGroup permanently destroys all data in it.

Volume

A Volume is a named data container within a VolumeGroup. Clients mount or access a Volume after it reaches Ready phase.

Volume Types

Type	Client interface	Common workloads
POSIX	FUSE filesystem mount	General filesystem access, AI/ML datasets
HDF5	Native HDF5 access	Scientific data in HDF5 format
MPIIO	Native MPI-IO access	HPC simulation output
PYTHON	Python client library	Python data pipelines

Snapshots

Volumes support point-in-time snapshots. Snapshots can be:

• Manual: Created on demand via the Dashboard or API
• Automatic: Configured via snapshotPolicy in the Volume spec

Backups

Volumes can be exported to an external location (S3, NFS) using the backupConfig spec field. Backups can be full or incremental, scheduled on a cron, hourly, daily, or weekly basis.