Chapter 34 Key Takeaways: COBOL-to-Cloud Patterns
Threshold Concept
Cloud is an architecture decision, not a destination. The question isn't "how do we move COBOL to the cloud?" The question is "which workloads benefit from cloud characteristics (elasticity, pay-per-use, geographic distribution) and which require mainframe characteristics (throughput, latency, data locality, five-nines)?" Organizations that treat cloud migration as a destination ("we need to get off the mainframe") make expensive mistakes. Organizations that treat it as an architecture optimization ("we need each workload on its best platform") succeed.
What Works on Cloud
| Workload | Why It Works | Example |
|---|---|---|
| Dev/test | Eliminates MIPS, fast provisioning, team isolation, modern tooling | Pinnacle: $1.8M/yr MIPS savings, 90-minute provisioning |
| Batch reporting | Elastic compute, periodic execution, read-only, latency-tolerant | CNB: 47 regulatory reporting programs on AWS |
| Analytics | Read-only, batch-fed, benefits from cloud ML/BI tools | Pinnacle: CDC-replicated claims data for fraud detection |
| Document generation | Batch, output-only, no system-of-record updates | Pinnacle: 2.4M letters/month, COBOL + Python PDF pipeline |
| Low-volume CICS (<100 TPS, >50ms acceptable) | Standard CICS API, self-contained, no cross-region communication | SecureFirst: 12 admin transactions, 200/day |
What Does NOT Work on Cloud
| Workload | Why It Doesn't | Hard Limit |
|---|---|---|
| High-volume OLTP (>500 TPS) | CICS emulation overhead, network latency, no Coupling Facility scaling | Throughput ceiling well below z/OS |
| Sub-10ms latency transactions | Every DB2 call crosses a network; the latency tax is 5-10x | Physics, not tuning |
| Two-phase commit (DB2 + MQ) | Cloud CICS emulators don't fully support distributed 2PC | Architectural gap |
| CICS MRO/ISC | Multi-region operation not replicated by rehosting platforms | Not available |
| Parallel Sysplex data sharing | No cloud equivalent to Coupling Facility | Fundamental: shared-everything vs. shared-nothing |
| Full-function CICS (events, channels, CICSPlex SM) | Emulation coverage is partial | 5% gap is where your critical code lives |
Rehosting Platforms at a Glance
| Platform | Approach | CICS Support | Best For | Watch Out For |
|---|---|---|---|---|
| Micro Focus Enterprise Server | Compiles COBOL to x86 native | Good (basic-moderate) | Broad compatibility, mature ecosystem | Per-core licensing adds up; packed decimal is software-emulated |
| Heirloom Computing | Converts COBOL to Java bytecode | Limited | No runtime license; deploys on any JVM | Converted code unreadable to both Java and COBOL developers |
| NTT DATA UniKix | CICS/batch emulation on Linux | Good (deepest CICS coverage) | CICS-heavy workloads | Smaller ecosystem; no IMS; DB2 migration required |
The 95% Trap: The 5% of COBOL that doesn't work on any rehosting platform is the hardest, most business-critical, least-tested code.
Data Synchronization Patterns
| Pattern | Latency | Complexity | Cost | Use When |
|---|---|---|---|---|
| Batch Extract | Hours | Low | Low | Reporting, analytics, regulatory snapshots |
| CDC | Seconds-minutes | Medium | Medium | Dashboards, near-RT analytics, fraud detection |
| API Real-Time | Milliseconds | High | High (MIPS) | Mobile/web, account services, user-facing interactions |
Rule of thumb: Start with batch extract. Move to CDC only when the business demands it. Use real-time API only for user-facing interactions where stale data is unacceptable.
The Hybrid Architecture
MAINFRAME: System of record, OLTP, core batch, five-nines workloads
CLOUD: Reporting, analytics, dev/test, API gateway, ML/AI
BETWEEN: Batch extract + CDC + z/OS Connect APIs
Lisa Tran's summary: "The mainframe does the work. The cloud does the looking."
TCO Reality Check
The Three Lies in Vendor TCO Slides
- Allocated cost =/= marginal cost. Removing 400 off-peak MIPS doesn't save $2M. It may save $0 if the R4HA peak is during online hours.
- Migration project costs are underestimated. Vendor estimates cover the easy 80%. The hard 20% (GDGs, SORT exits, REDEFINES, EBCDIC edge cases) doubles the cost.
- Ongoing operational costs are omitted. Cloud ops staff, Micro Focus maintenance, Direct Connect, security compliance, DR, monitoring — $200K+/year that's not in the vendor's number.
Kwame's Rule of TCO
"If your honest TCO shows less than 30% savings after including every cost you can think of, don't do it." - 50%+ savings: probably real - 30-50%: danger zone — is the operational risk worth the marginal savings? - <30%: don't do it — hidden costs will eat the margin
The MIPS-to-Dollars Fallacy
The "$5,000/MIPS" number is the fully allocated cost. The marginal cost of removing a specific workload depends on whether that workload runs during the R4HA peak. For off-peak batch, the marginal cost is often zero or negative (removing off-peak work can increase per-MSU pricing for remaining workloads).
The Decision Framework
Three Questions
- What is the workload's primary characteristic? (Batch/OLTP/dev-test/Sysplex-dependent)
- What is the marginal mainframe cost savings? (Not allocated — marginal)
- Can you operate it? (Do you have the people to run COBOL on cloud at 2am?)
The Decision Tree
Sysplex data sharing? → STOP. Mainframe.
High-volume OLTP? → STOP. Mainframe.
Dev/test? → MOVE TO CLOUD.
Batch reporting? → EVALUATE (marginal cost + non-cost benefits)
Low-volume CICS? → EVALUATE CAREFULLY (standard API only)
Everything else? → DEFAULT TO MAINFRAME.
The Cloud Migration Maturity Model
| Stage | Timeframe | Focus | Key Deliverable |
|---|---|---|---|
| Cloud Curious | 6-12 months | Dev/test, team skills, connectivity | Landing zone, first dev/test environments |
| Cloud Capable | 12-24 months | First production batch, CDC | First production workload, validated TCO |
| Hybrid Optimized | 24-48 months | Multiple workloads, integrated ops | Documented architecture, runbooks |
| Strategic Hybrid | 48+ months | Routine placement decisions | Mature operations, optimized TCO |
Key Numbers from the Chapter
- 2:40 → 6:15 — CNB's regulatory batch window expansion (mainframe → cloud) due to I/O architecture difference
- 87% → 55% — CNB's vendor-claimed vs. honest TCO savings
- **$0/year** — CNB's mainframe savings from removing off-peak batch (allocated cost was $2M)
- +$84K/year — CNB's mainframe cost increase from removing off-peak batch (pricing model impact)
- 8.4ms → 340ms — FBA eligibility p99 latency (mainframe → cloud) due to buffer hit ratio drop
- $43M spent, $75M projected — MidWest Mutual's failed full migration (Case Study 1)
- $730K invested, $1.46M/year savings — Pinnacle Health's successful hybrid (Case Study 2)
- 19.4% — Pinnacle's MIPS reduction (7,200 → 5,800) through targeted cloud migration
Rob Calloway's Final Lesson
"Analytics warehouse. Same pattern as reporting — read-only, batch-fed, elastic demand. And this time, I know what to budget."
That's what progress looks like. Not a PowerPoint slide with two boxes and an arrow. A spreadsheet with real numbers, a team with real experience, and a decision made with real data.