Hyperconverged Infrastructure and Distributed Storage Concepts
This domain tests how the Distributed Storage Fabric (DSF) moves and protects data: the I/O path, redundancy, rebuild behavior, and space-efficiency features.
The I/O path
- Every node runs a Controller VM (CVM). Guest writes go over the network to a CVM - VMs never touch physical disks directly.
- Stargate handles I/O and the extent store; Curator performs disk balancing and background scans; Cassandra holds cluster metadata; Genesis supervises services.
- A write is acknowledged only after all replicas are persisted. Data-locality heuristics keep reads on the local CVM where possible, which is why the network is the performance backbone.
Resilience
- RF2 keeps two copies (tolerates one failure); RF3 keeps three (tolerates two). Raising the redundancy factor requires free space.
- Capacity is reserved for rebuilds. Checksums and scrubs detect and repair corruption. Erasure Coding (EC-X) cuts overhead for colder data and requires RF2 or higher.
- Deduplication and compression can run inline or post-process, configured per storage container.
Worked example
Five nodes with 8 TB drives give 40 TB raw; RF2 gives about 20 TB usable. If 12 TB is stored and one node fails, 32 TB raw remains. Re-protecting 12 TB at RF2 needs 24 TB, which fits - so a complete rebuild is possible. If 18 TB were stored, the requirement would be 36 TB against 32 TB available, and the cluster would stay unprotected until capacity is added or the node returns.
Traps
Confusing raw and usable capacity; believing EC replaces RF; assuming rebuild is instantaneous or free of I/O impact; assuming dedupe is always enabled on every container.