Skip to main content

Overview

XSDS offers two data protection strategies: replication and erasure coding. The appropriate choice depends on your durability requirements, acceptable storage overhead, and the latency sensitivity of your workload.
Prerequisites
  • An active Xloud account with project member access
  • Contact your storage administrator to provision erasure-coded pools — they are configured at the cluster level by an administrator. Your administrator can configure this through XDeploy.

Protection Strategy Comparison

StrategyStorage OverheadLatencyFailure ToleranceBest For
Replication (factor 3)3× raw capacityLow2 simultaneous OSD failuresDatabases, boot volumes, latency-critical
Erasure Coding (4+2)1.5× raw capacityHigher2 OSD failuresBackups, archives, large object stores
Erasure Coding (8+3)1.375× raw capacityHigher3 OSD failuresVery large datasets requiring maximum efficiency

Replication

Replication stores multiple complete copies of every object across separate OSDs. The default replication factor is 3 — three full copies are maintained at all times.
When data is written to a replicated pool, XSDS simultaneously writes it to the number of OSDs defined by the replication factor. Each copy is placed on a different host according to the CRUSH map, ensuring that host-level failures do not cause data loss.
Replication FactorCopiesSurvives Simultaneous FailuresStorage Overhead
22 copies1 OSD failure
3 (default)3 copies2 OSD failures
44 copies3 OSD failures
Characteristics:
  • Lower read/write latency compared to erasure coding
  • Higher raw storage overhead (3× capacity for factor-3)
  • Recovery is faster after an OSD failure
  • Best suited for latency-sensitive workloads (databases, boot volumes)

Erasure Coding

Erasure coding divides data into data chunks and parity chunks, distributing them across OSDs. The data can be reconstructed from any sufficient subset of chunks, providing durability at significantly lower storage overhead than replication.
In an erasure-coded profile k+m, data is split into k data chunks and m parity chunks. The pool can tolerate the loss of any m OSDs and reconstruct the data from the remaining k chunks.
ProfileData Chunks (k)Parity Chunks (m)OverheadFailure Tolerance
4+2421.5×2 OSD failures
6+2621.33×2 OSD failures
8+3831.375×3 OSD failures
Characteristics:
  • Significantly lower storage overhead compared to replication
  • Higher CPU and network overhead for encoding/decoding
  • Best suited for large objects: backups, archives, cold data
  • Recovery after failure takes longer than replication

Validation

Confirm your volume is using the expected protection policy:
Navigate to Project → Volumes → Volumes and click on the volume name. The Volume Type field shows which pool backs this volume. Contact your storage administrator to confirm the protection scheme for that pool type. Your administrator can configure this through XDeploy.

Best Practices

Production Workloads

Use replication factor 3 for all production databases, application volumes, and frequently-accessed data. The lower latency outweighs the higher storage cost for latency-sensitive workloads.

Backup & Archive

Use erasure-coded pools (4+2 or 6+2) for backup targets and archives. The storage efficiency savings are significant at scale, and higher latency is acceptable for infrequent access patterns.

Development Environments

Replication factor 2 is sufficient for non-critical development and test volumes. Reduces storage consumption while maintaining single-failure protection.

Pre-Maintenance Snapshots

Combine data protection pools with regular snapshots for layered protection. Pools protect against hardware failures; snapshots protect against logical corruption and accidental deletion.

Next Steps

Snapshots

Create point-in-time snapshots to complement pool-level data protection

Storage Tiers

Route workloads to NVMe, SSD, or HDD tiers based on access patterns

XSDS Admin — Pool Management

Create and configure replicated and erasure-coded pools (administrator)

XSDS Admin — CRUSH Maps

Configure failure domains and device class rules (administrator)