Hybrid Cloud Burst¶

Service ownership

Owner: dc-operations (colo-pm@clouddigit.ai) — Status: GA — Last audited: 2026-05-11

Connect a colocation footprint into Cloud Digit cloud over private fabric — keep the colo as steady-state baseline, burst into cloud for peak.

What it is¶

A networking pattern (and a billing arrangement) that lets you treat your colo cabinets as the primary production environment and Cloud Digit cloud as elastic spillover. Traffic is routed over a private cross-connect, so it never traverses the public Internet.

Topology¶

graph LR
    subgraph "Colo (your steady state)"
      Colo1[App tier]
      Colo2[DB primary]
    end
    subgraph "Cloud Digit cloud"
      LB[Load Balancer]
      ASG[Auto Scaling Group]
    end
    Internet((Internet)) --> LB
    LB --> ASG
    ASG -->|private fabric| Colo1
    Colo1 -->|sync replication| Colo2
    ASG -.peak only.-> Colo2

Use cases¶

Holiday / event-day surge capacity (cricket finals, Eid, election results)
New-product launch traffic
Periodic batch (monthly close, end-of-quarter reporting) without permanent cloud footprint
DR — cloud as the fallback target when colo is unavailable (alternative pattern: see DRaaS)

Network configuration¶

Cross-connect between your cage/rack and Cloud Digit cloud fabric
VPC in the cloud side, peered into your colo VLANs via the cross-connect
Routing — BGP-default; static routes possible for simple cases
Security — security groups on the cloud side; you operate firewalls on the colo side
DNS — split-horizon supported; cloud-side resolver can serve colo records

Cost story¶

Steady-state stays in colo (predictable monthly spend)
Peak burns cloud per-hour
Cross-connect is a fixed monthly fee
Net effect: most months you spend the colo flat fee; peak months spend extra cloud

Pricing¶

Cross-connect at standard rates
Cloud resources at standard hourly / per-second rates

See Pricing.

Operate this service¶

AdministrationOperationTroubleshooting

Connect your colo deployment to Cloud Digit cloud services for burst capacity, DR, or hybrid architectures.

When this fits¶

Steady-state on colo gear, peaks burst to cloud
Colo as primary, cloud as DR
Cloud-native services (managed DB, AI) alongside colo'd legacy gear
Migration in progress (colo → cloud, gradual)

Topology¶

[Colo cage / racks] ↓ cross-connect [Cloud Digit VPC] ↓ [Cloud services: VMs, K8s, managed DBs, etc.]

The cross-connect makes the colo network indistinguishable from a VPC subnet — same IPs, same security groups, same routing.

IAM¶

Standard CD IAM applies — cloud-side resources governed by cd.* roles; colo-side by colo.* roles. Cross-cutting concerns (network policy, audit) integrate.

Network design¶

Colo network as a first-class VPC subnet: - Allocate CIDR from your VPC supernet (e.g., 10.50.0.0/16 for colo, 10.0.0.0/16 for cloud) - Routes between colo and cloud subnets implicit in the VPC - Security groups apply uniformly - VPC Flow Logs capture all cross-domain traffic

Burst patterns¶

Pattern	Use
Cloud-burst compute	Colo for baseline, cloud Auto Scaling for peaks
Cloud-burst storage	Colo for hot data, cloud S3 for cold/overflow
DR-burst	Colo primary, cloud cold standby ready to scale
Workload migration	Gradual move; eventually decom colo

Cost¶

Cross-connect: monthly fee per fiber pair. Burst usage: standard cloud rates. The hybrid premium is small; the benefit is large.

Related¶

Metrics¶

Cross-domain visibility:

Metric	Notes
`hybrid.colo_to_cloud.bytes_per_sec`	Inter-domain traffic
`hybrid.cloud_to_colo.bytes_per_sec`	Reverse direction
`hybrid.cross-connect.utilization_pct`	Capacity planning
`hybrid.burst.cloud_instances`	Cloud capacity in use
`hybrid.failover.last_drill_age_days`	DR exercise cadence

Burst trigger patterns¶

Auto-burst: - Auto Scaling Groups in cloud configured with hybrid-burst-on-trigger - Trigger: colo capacity threshold (e.g., > 85% sustained for 10 min) - Cloud ASG scales out; LB distributes traffic across colo + cloud

Manual burst: - Operator decides; pre-provisioned cloud capacity scaled up

Workload portability¶

Design workloads to run identically in colo and cloud: - Use container images (run anywhere) - Avoid hardware dependencies (custom NICs, GPU-direct) - Externalize state (managed DBs, S3)

If workload is colo-only by design (e.g., uses colo'd FC SAN), it can't burst.

Cost optimization¶

Cloud burst is expensive per hour but cheap overall (only pay during peaks): - For frequent peaks: increase colo baseline - For rare peaks: keep colo lean, burst more - Right balance: 6-month cost analysis

DR pattern¶

Colo + cloud DR: 1. Continuous replication (DBs, files, configs) to cloud 2. Cold standby cloud resources 3. Periodic DR drill (failover, validate, fail back) 4. Real failover triggered manually or by health-check

See DRaaS for orchestration.

Related¶

Burst not triggering¶

ASG configured to burst on colo capacity threshold, but isn't scaling: - Verify trigger metric is firing (cd colo metric --rack <id> --metric capacity_pct) - ASG max instance limit reached - Cross-connect capacity insufficient (burst traffic + existing = saturated) - Cloud VPC capacity in target region

Cross-connect saturated¶

hybrid.cross-connect.utilization_pct > 95%: - Real burst happening (good — order more cross-connect capacity) - Asymmetric traffic (one direction hot, other quiet) — investigate - Bug: cloud-burst egressing through cross-connect when cloud-cloud path is shorter

Latency higher than expected¶

Cross-domain (colo ↔ cloud) latency: - Same DC: < 1 ms - Different DC (intra-BD): 5-15 ms

If observing 30+ ms intra-DC: ticket. Likely routing issue.

Workload doesn't work in cloud¶

Bursted to cloud, but the workload fails: - Hardware dependency (FC, custom NIC) - Stateful and assumes local storage - License-tied to specific physical CPU - Network ACLs allow colo but not cloud subnets

Fix at workload design layer; not all workloads are burst-ready.

DR failover failed¶

DR drill exposed issues: - Replication caught up but gap discovered (manual data not replicated) - DNS cutover incomplete - Cloud capacity insufficient (didn't pre-warm) - Cross-cutting auth tokens region-locked

Document, fix, re-drill. The whole point of drills is to find these.

Bill spike from burst¶

Burst hours add up at cloud rates: - Was the burst justified (real demand) or runaway? - ASG max instances cap; verify reasonable - For sustained higher baseline, increase colo capacity instead

Hybrid network segmentation issues¶

Workloads in cloud can't reach colo'd services: - Routes between subnets configured? - Security groups allow cross-domain traffic? - DNS resolution works (private hostnames need to resolve from both sides)?

bash cd network reachability test --from cloud-vm-01 --to colo-server-04