Skip to content

Adaptive Storage Architecture

Balancing storage performance and cost is a core proposition of cloud architecture. Cyberun Cloud abandons the "one-size-fits-all" storage approach and implements a Scenario-Based Tiered Storage Strategy, providing precisely matched IOPS and durability guarantees for different business needs.

Storage Class Definitions

We have predefined standardized StorageClasses at the platform level, allowing developers to declare them directly based on business requirements.

Tier 1: Extreme Performance (sc-nvme-local)

  • Technical Implementation: Local Direct-Attached NVMe SSD (PCIe Gen4).
  • Performance Metrics: Latency <0.05ms | IOPS 100K+.
  • Use Cases: AI training scratchpads, high-frequency trading caching, ephemeral data processing.
  • Architectural Trade-off: Provides physical limit I/O speeds. Data is managed with the node lifecycle, suitable for stateless or application-layer replicated high-performance workloads.

Tier 2: Strong Consistency Redundancy (sc-longhorn-rep3)

  • Technical Implementation: Longhorn-based hyper-converged storage using Synchronous Replication.
  • Performance Metrics: Latency ~0.5ms | Data Durability 99.999%.
  • Use Cases: Kubernetes metadata (etcd), distributed databases (StatefulSets), message queues.
  • Architectural Advantage: Write acknowledgment mechanism ensures all replicas are persisted before returning success, providing strong consistency guarantees.

Tier 3: Elastic Disaggregated Storage (sc-ceph-rbd)

  • Technical Implementation: Rook-Ceph based external storage cluster mounted via high-speed network.
  • Performance Metrics: Latency ~1.0ms | Capacity PB-scale Scaling.
  • Use Cases: Massive data archiving, shared file systems (RWX), backup snapshots, data lakes.
  • Architectural Value:
  • Independent Scaling: Decouples compute resources (CPU/RAM) from storage resources (Disk). When data volume surges, there is no need to pay for idle compute power, significantly reducing Total Cost of Ownership (TCO).
  • Failure Isolation: Even if the compute cluster suffers a catastrophic failure, data in the storage cluster remains intact and can be immediately mounted to a standby cluster.

Data Protection and Lifecycle

  • Snapshot-as-a-Service: Integrates CSI snapshot standards, supporting Application-Consistent Snapshots to ensure logical integrity of databases during backup.
  • Offsite Disaster Recovery: Core data volumes enable asynchronous replication strategies by default, mirroring data increments to offsite data centers, meeting enterprise-grade RPO (Recovery Point Objective) requirements.