Skip to main content

Deployment Models

Scalability​

Platform Scale
  • GPU Nodes — Scalable to thousands of nodes per deployment
  • Accelerators — Scalable to thousands of GPUs per deployment
  • GPU Vendors — NVIDIA, AMD, Intel multi-vendor support
  • Tenants — Multi-tenant with dedicated IAM realms per tenant
  • Clusters per Tenant — Configurable via quota system
  • Network Isolation — Configurable pool of tenant VRFs, multiple VLANs per VRF, expandable

Tenant Onboarding Flow​

  1. Domain Created — Platform Super Admin creates tenant with name, initial quota, and network allocation
  2. Identity Provisioned — IAM realm auto-created with default roles, clients, and auth flows
  3. First Admin User — Domain Admin created with full tenant-level permissions
  4. Network Allocated — VRFs/VLANs assigned (multiple VLANs per tenant), resource quotas configured
  5. Organization & Project Setup — Domain Admin creates org/project hierarchy, sets per-project quotas
  6. Users Invited — Users created/invited with appropriate roles and project assignments
  7. Ready — Tenant fully operational. Can provision bare metal, create clusters, deploy workloads

Support & SLA​

High Availability Infrastructure​

The control plane utilizes multiple redundancy strategies:

  • Portal & Load Balancing — Multiple replicas with automatic failover
  • Identity Management — Clustered IAM with session replication
  • Database Layer — Primary-replica configuration with rapid failover
  • Service Coordination — etcd quorum for distributed consensus
  • Metrics — HA metrics cluster with data replication

Data Protection​

  • Configuration Backups — Automated backups every 6 hours
  • Retention — 12-24 months in object storage
  • Audit Logs — Continuous storage with long-term retention
  • Metrics — 90 days hot storage, 12 months cold archives

Monitoring & Alerts​

Comprehensive monitoring with exporters covering every infrastructure layer. Critical thresholds trigger alerts:

  • GPU temperature exceeding safe limits
  • CPU utilization above threshold
  • Memory utilization at node level
  • Storage latency anomalies

Incident Management​

Incidents follow severity classifications from Critical (platform-wide outages) through Low (minor issues with workarounds). Response includes detection, triage, investigation with correlation ID tracking, resolution, and post-incident review.

Support Tiers​

  • Standard — Business-hours ticket support with next-business-day response
  • Enterprise — Priority response for critical issues plus dedicated engineer
  • Premium — Rapid critical response with assigned team and custom SLA negotiation

Specific SLA terms, response times, and support tiers are defined per customer engagement.