KT Cloud shifts from public-sector DR backup focus to tiered A1–A4 recovery architectures

Public-sector DR is transitioning from simple data backup to full service recovery, with architectures proposed based on A1–A4 tier RTO and RPO standards
Tier A1 combines Seoul Multi-AZ Active-Active with Gyeongbuk region DR, while Tier A4 applies remote dispersed backup and Cold DR
DR for AI workloads centers on securing training checkpoints and GPU resources, requiring distinct recovery strategies for inference and training
KT Cloud PLATFORM is advancing standard architectures, transition automation, Seoul–Gyeongbuk wide-area DR, and AI recovery support features

Recovery Objectives and Architecture

Tier A1 targets RTO within 1 hour and RPO of 0 (real-time), combining Active-Active with remote DR
Tier A2 targets RTO of 3–12 hours with Hot Standby, Tier A3 uses Warm Standby or Pilot Light, and Tier A4 uses Cold DR
To optimize cost efficiency, resources are minimized during normal operations, with IaC-based automatic scaling and pre-secured DB and storage capacity during failures

Multi-AZ addresses data center failures, Multi-Region addresses wide-area disasters, and GSLB handles traffic switching
Tiers A1 and A2 require one live switchover per year, Tier A3 requires tabletop and simulation exercises, and Tier A4 mandates tabletop exercises including remote recovery
Verifying data reverse replication and primary center restart procedures is essential during failback

AI inference maintains minimum capacity across multiple regions and scales during failures, while training resumes based on checkpoints
Recovery scope includes training data, model registries, GPU resources, and pipeline environments
To reduce GPU costs, resource allocation based on workload criticality and securing alternative resources are necessary