IBM Fusion: Four Years Behind the Air Gap

Designing, running and recovering applications within an air-gapped IBM Fusion platform on bare metal OpenShift. The case studies below follow the work from design to recovery.

I led the design of an air-gapped platform on IBM Fusion HCI and bare metal Red Hat OpenShift, and lead the in-house team that runs and upgrades it. It sits inside a high-security, air-gapped data centre. Every component arrives through an approved supply chain, nothing can be fetched on the day, and change is measured in months, not days, so the platform has to absorb failures on its own and every upgrade has to be planned before a single image is mirrored.

Over four years the work moved from architecture and deployment to operating, recovering and improving the platform. We scaled out across eight medium nodes rather than four dense ones, made upgrades a planned, repeatable practice run by our own people, built observability that took the cluster from stable to understood, gave application teams a secure way to build and release software inside the air gap, and proved that applications can be restored, not just backed up. When Red Hat changed its bare metal subscription, I re-ran the business case.

The case studies below each follow one problem through its constraints, the decision we made and the outcome.

Case studies

Problem → constraints → decision → outcome

Case study 01 · Platform reliability

Know Your Cluster

“It’s green, so it must be healthy.”

Most teams treat a green dashboard as proof of health and only look deeper once something breaks. We built the observability to see what self-healing hides, and the discipline to act on it before it became an outage.

  • 30+ dashboards
  • 0 chronic alerts
Read the case study: Know Your Cluster
is it up? healthy? tuned?

Case study 02 · Platform lifecycle

Keeping pace

“How do we keep pace with upgrades?”

Modern platforms release continuously, and a high-security environment can’t download its way to the next version. We made upgrades a planned, repeatable practice, run by our own people and kept inside vendor support.

  • Air-gapped
  • Run in-house
Read the case study: Keeping pace
connected air gap upgrade upgrade next one

Case study 03 · Platform architecture

Scale Out versus Dense

“Custom-built looks cheaper on paper.”

Four dense nodes would have been compact, but losing one takes away a quarter of the cluster. Behind an air gap, where help can’t be downloaded, we chose eight medium nodes on bare metal OpenShift and IBM Fusion HCI, and kept running when one failed.

  • 8 medium nodes
  • Air-gapped
  • Bare metal
Read the case study: Scale Out versus Dense
dense: −25% scale out: −12.5%

Case study 04 · Backup and recovery

Db2 Backup & Restore

“The sample assumes one namespace. Production has more.”

A backup is only as good as its restore. IBM’s sample Db2 recipe assumed the operator and the database share a namespace; ours don’t. I adapted it so each database is backed up on its own terms and can be restored into another namespace.

  • Backup & restore
  • Cross-namespace restore
  • Open-source recipe
Read the case study: Db2 Backup & Restore
restore production copy, elsewhere

Case study 05 · Developer platform

Air-Gapped Internal Developer Platform (IDP)

“No internet, and still expected to be agile.”

Inside an air gap, developers can’t pull a package or an image from the internet. Four operators on a local registry give a small team reproducible workspaces, automated quality gates and image scanning, from commit to release.

  • Secure by design
  • Air-gapped
Read the case study: Air-Gapped Internal Developer Platform (IDP)
codetestscan cluster blocked nothing fetched from outside

Case study 06 · Business case

Same node, new price

“Is bare metal still the cheaper option?”

We chose bare metal OpenShift partly on licensing. When Red Hat moved to one subscription per node, I re-ran the comparison at public list prices and proposed a revision of the IBM Fusion product guide.

  • Business case
  • Public list prices
Read the case study: Same node, new price
thennow ? core pairs per node

Running IBM Fusion or OpenShift behind an air gap? Let’s compare notes.

Connect on LinkedIn