🛒 Globex E-Commerce — Animated Case Study

What happens when you click

Scrolling reveals each chapter (IntersectionObserver) and triggers its SVG/CSS animation — e.g., the monolith blocks crumble, the dependency lines draw, traffic flows to the cloud.
Click any SVG component (Oracle, SOAP pricing, Exchange relay, containers, CSPM findings) to open its explanation in the detail box under that chapter.
▶ Replay story — GET api/globex fetches the nine chapters from the backend and plays them through the flow + timeline at the bottom.
Team chips filter the chapter list to one team's responsibilities.

1 · Business context

Globex sells 40,000 SKUs through a single Java WAR deployed on 40 VMware ESXi hosts. Black Friday 2024 went down for 3 hours — $2M per hour lost. The data-centre lease expires in 18 months and the CFO refuses to renew.

Decision

Exit the data centre; do not rewrite the platform before the move.

2 · BSA dependency discovery

Four BSAs interview 14 application owners and run Cloudamize agents for 30 days. The map shows the monolith calling Oracle, a legacy SOAP pricing service owned by Finance, and an on-prem Exchange relay for order emails — none of which were in the original scope.

Click a component in the diagram.

3 · Decision: strangler pattern

SteerCo approves a hybrid plan: Rehost the monolith with AWS MGN in wave 2, Replatform Oracle to RDS, and carve out Catalog and Checkout into containers after landing — the strangler-fig pattern. Refactoring everything first was estimated at 30 months, 12 past the lease.

Why this works

Risk moves off the lease date immediately; modernisation happens where compute is elastic and experiments are cheap.

4 · Developers containerise

Developers wrap the WAR in a Docker image, externalise configuration to environment variables, remove local-disk session state in favour of ElastiCache, and build a CI/CD pipeline that pushes to ECR. Connection strings change from 10.20.4.17:1521 to orders.cluster-xyz.rds.amazonaws.com.

Click a stage.

5 · QA baseline

Before anything moves, QA records a baseline: p95 checkout 420 ms, 1,200 orders/min at peak, 0.4% error rate. Without this, “the cloud is slower” is an opinion; with it, it is a measurable gate.

6 · Cutover weekend

Friday 20:00 traffic stops; Saturday 02:00 the final DMS delta lands; QA verifies row counts (48.2M orders both sides); Saturday 10:00 Route 53 flips the CNAME. Cloud p95 is 310 ms — within the 300 ms gate? No: 10 ms over. The SteerCo chair accepts the variance in writing because autoscaling had not yet warmed; by 11:00 it is 280 ms.

7 · Ops: CSPM after migration

Within 48 hours, Cloud Security Posture Management (AWS Security Hub / Microsoft Defender for Cloud) raises findings that never existed on-prem. Click each.

Click a finding.

8 · Outcome & lessons

Black Friday 2025: 3× traffic, zero downtime, autoscaling from 6 to 40 pods. Run-cost down 28% after right-sizing and Savings Plans. Lease exited on time.

Rehost first, refactor later.
Baseline before you move.
Security parity is a gate, not an afterthought.
Hidden dependencies are found by network flows, not interviews alone.

Replay from the backend

AllBSADevelopersQAOps

Context

➜

Discover

➜

Prepare

➜

Cut over

➜

Operate

Click a node to learn what it does.

Chapters from the API appear here.

Deep dive

How

Discovery → strangler plan → containerise → baseline → cutover → CSPM.

Why

Lease deadline forces sequencing; revenue risk forces measurement.

When

Any monolith with a hard exit date and revenue exposure.

Pitfalls

Lift-and-shifting “allow any” firewall rules.
Forgetting external allow-lists (payment gateway IPs).
No baseline → arguments instead of decisions.

Key ideas

Strangler figDependency mapBaselineCSPMHypercare

🐍 Python data served (GLOBEX chapters)

Static list sim.GLOBEX — fetched via GET api/globex (see output above).