CIOPages

CIOPages · Trench Notes · Infrastructure

Modernizing mainframe applications

The work of modernizing mainframe applications, on the platform or off it: the options, the order of the work, the risks to plan for, the IBM software bill.

Published November 17, 2026

What the job is

Mainframe modernization means changing how mainframe applications are built, run or paid for. Some programs move every application off the mainframe. Others keep the mainframe and change the applications on it.

Triggers include a hardware or software contract renewal, a shrinking pool of staff who know the code, and pressure to connect old systems to new ones.

AWS, Google Cloud and IBM all offer AI tools that analyze COBOL, document it and, in some cases, convert it. AWS has closed its AWS Mainframe Modernization service to new customers, in both its managed and self-managed forms. Existing customers can keep using it. AWS points new projects to AWS Transform and to offerings bought directly from vendors.

Off the mainframe, or modernized on it

Decide per application, not per mainframe. Give each application one of six labels: keep and modernize in place, rehost, refactor, rewrite, replace with a package, or retire.

Leaving the mainframe completely removes the mainframe cost only when the last application leaves. Until then, both platforms run and both are paid for. Model the cost of that transition period, including whether the applications that leave change the IBM software bill.

Consider leaving some applications where they are: those with heavy transaction volumes and tight response times, those nearing replacement by a package, and those whose output is checked by a regulator and would need full revalidation.

The options

OptionWhat it involvesExample tooling
Modernize in placeKeep the application on z/OS. Expose it through APIs and improve the code there.IBM z/OS Connect (REST APIs to and from z/OS), IBM watsonx Code Assistant for Z
Rehost or replatformRecompile COBOL and PL/I largely unchanged and run it on a mainframe-compatible runtime on Linux, Windows or cloud servers.Rocket Software (formerly Micro Focus) Enterprise suite
Automated refactorConvert the code to Java or .NET, and hierarchical or network databases to relational ones.AWS Transform for mainframe (COBOL to Java)
RewriteExtract the business rules and requirements, then build a new application from them.AWS Transform business logic extraction and requirements generation
ReplaceMove the business process to a commercial package or SaaS product.None for the code
RetireSwitch off applications nobody uses, after archiving the data that must be kept.None

Microsoft's Azure architecture guidance describes two ways to run a refactor: convert the whole mainframe in one process, or move applications gradually. The first removes interim mainframe maintenance and facility costs, but conversion, data migration and testing must all align. The second saves cost application by application and lets each conversion inform the next.

The order of the work

  1. Inventory the code. Collect every COBOL, PL/I and Assembler program, copybook, JCL job, CICS and IMS definition, and database definition. Compare what you collected with what runs in production. AWS Transform lists missing files after its code analysis. Google's Mainframe Assessment Tool scans COBOL, JCL, Assembler and other code and generates documentation.
  2. Map the data and the interfaces. List every Db2 table, IMS database and VSAM file, and every file transfer, extract and message queue to and from other systems.
  3. Establish the cost baseline. Get the IBM and third-party software contracts, their renewal and notice dates, and the monthly usage reports IBM bills from.
  4. Label every application. Keep, rehost, refactor, rewrite, replace or retire, with a business owner for each decision.
  5. Build the test approach before converting anything. Capture production inputs and outputs that the new system must reproduce.
  6. Pilot one application end to end. Choose one with batch and online parts. Measure conversion effort, defects found in testing and run times against the mainframe.
  7. Migrate in waves, with parallel running. Run old and new side by side and compare outputs until the business accepts the results.
  8. Decommission. Remove the workload from the mainframe, then reduce capacity and software licenses at the next opportunity in the contract.

Risks to plan for

  • Finding the source. Check that source exists for every program that runs in production, and that it matches the running module. AWS Transform reports missing files and missing references found in its code analysis.
  • Data migration. Mainframe files use EBCDIC encoding, which Linux and Windows systems do not read directly. Conversion follows the record layouts in the copybooks. Moving IMS or VSAM data to a relational database also changes its structure. Test each conversion rule against real data.
  • Proving equivalence. The new system must produce the same results as the old one, to the cent, across month-end, year-end and error cases. Google Cloud's Dual Run runs workloads on the mainframe and on Google Cloud at the same time so their outputs can be compared. Its documentation lists batch workloads as supported and online transactions as in preview.
  • Batch run times. Overnight batch has to finish inside the same window on the new platform. Run a full-volume test to confirm it does.
  • Readable converted code. AWS documents that its first COBOL-to-Java pass keeps COBOL-style data structures and variable names. It offers a second step, Reforge, to restructure the code into conventional Java. Plan for the team that will maintain the result to review it.

Details to put in the plan

  • The batch scheduler and its job dependencies, which carry business logic that is not in any program.
  • Sort, print and output management utilities that jobs call.
  • Security rules held in the mainframe security product, which need an equivalent on the target.
  • Every downstream system that receives a mainframe file and expects its exact layout.
  • Performance records. AWS Transform analyzes SMF (System Management Facility) records to show how batch jobs and CICS transactions are used. AWS recommends at least 13 months of records to capture annual events.
  • Data retention. Retired applications can hold records that must be kept and still be readable.

The IBM software bill and the contracts

Moving work off the mainframe does not lower the IBM software bill automatically. How much it lowers depends on the pricing model.

Under IBM's traditional sub-capacity pricing, a product's monthly charge is based on the highest four-hour rolling average utilization of the logical partitions where it runs, during the month. Moving an application that does not run during that peak period may not reduce the charge.

Tailored Fit Pricing has two software models. The Software Consumption Solution (formerly Enterprise Consumption Solution) charges on MSUs consumed, so reduced usage reduces the charge. The Enterprise Capacity Solution ties charges to the size of the physical environment, not to usage.

  • Which pricing model is each IBM product on, and when can it change?
  • Which third-party mainframe products have their own contracts, and what are their notice periods?
  • Can capacity be reduced in steps as waves leave, or only at renewal?
  • What will the target platform's software cost, including the emulation or runtime licenses?

AWS lists the mainframe modernization agent in AWS Transform as free on its pricing page, and says future capabilities may be paid. Testing, data migration and double running need their own budget lines.

Skills and operations

Keep the mainframe team who know the applications for the whole program: to explain the code, build test cases and run the old system during parallel running. Agree their role and retention early.

The target needs its own operations team, with batch scheduling, monitoring and recovery procedures that match what the mainframe provided. Rehearse a failure and a recovery on the new platform before the first production wave.

Settle these first

  • When do the IBM and third-party software contracts renew, and on which pricing model?
  • Do you have source code for every program that runs in production?
  • Which applications should stay on the mainframe, and who decided?
  • How will you prove the new system produces the same results as the old one?
  • Who will maintain the converted code, and in which language?
  • Who from the mainframe team will stay for the whole program?

Questions

Is AWS Mainframe Modernization still available?
Not to new customers. Both its managed and self-managed experiences are closed to new customers. Existing customers can keep using them. AWS points new projects to AWS Transform and to vendor-direct offerings.
What is the difference between rehosting and refactoring a mainframe application?
Rehosting recompiles the COBOL or PL/I largely unchanged and runs it on a mainframe-compatible runtime off the mainframe. Refactoring converts it to another language, such as Java, and can convert pre-relational databases to relational ones.
Will moving applications off the mainframe cut our IBM software bill?
It depends on the pricing model. Under traditional sub-capacity pricing, charges follow the peak four-hour rolling average utilization in the month, so moving work outside that peak may not reduce them. Under Tailored Fit Pricing's Software Consumption Solution, charges follow MSUs consumed.
Can generative AI convert our COBOL to Java?
Tools exist for it. AWS Transform refactors COBOL-based workloads into Java, and IBM watsonx Code Assistant for Z converts COBOL to Java. The converted code still needs equivalence testing and review by the team that will maintain it.
Sources. AWS, AWS Mainframe Modernization availability change: https://docs.aws.amazon.com/m2/latest/userguide/mainframe-modernization-availability-change.html; AWS, What is AWS Mainframe Modernization?: https://docs.aws.amazon.com/m2/latest/userguide/what-is-m2.html; AWS, AWS Transform: Modernization of mainframe applications: https://docs.aws.amazon.com/transform/latest/userguide/transform-app-mainframe.html; AWS, AWS Transform: Transformation of mainframe applications: https://docs.aws.amazon.com/transform/latest/userguide/transform-app-mainframe-workflow.html; AWS, AWS Transform pricing: https://aws.amazon.com/transform/pricing/; Microsoft, General mainframe refactor to Azure: https://learn.microsoft.com/en-us/azure/architecture/example-scenario/mainframe/general-mainframe-refactor; Microsoft, Mainframe file replication and sync on Azure: https://learn.microsoft.com/en-us/azure/architecture/solution-ideas/articles/mainframe-azure-file-replication; Google Cloud, Dual Run overview: https://docs.cloud.google.com/mainframe-dual-run/docs/dual-run-overview; Google Cloud, Mainframe Assessment Tool overview: https://docs.cloud.google.com/mainframe-assessment-tool/docs/overview; IBM, watsonx Code Assistant for Z: https://www.ibm.com/products/watsonx-code-assistant-z; IBM, z/OS Connect: https://www.ibm.com/support/z-content-solutions/zos-connect/; IBM, z/OS 3.1 Planning for Sub-Capacity Pricing: https://www.ibm.com/docs/en/SSLTBW_3.1.0/pdf/e0zi100_v3r1.pdf; IBM, Tailored Fit Pricing for IBM Z: https://www.ibm.com/support/z-content-solutions/tailored-fit-pricing/; Rocket Software, Enterprise Server data sheet: https://www.rocketsoftware.com/sites/default/files/resource_files/enterprise-server.pdf

Keep reading

Buyer Guide
AI Legacy Code Modernization
Checklist
Third-Party Risk Management Checklist

Assess and monitor vendor security and compliance risks.

Free checklist
Project Kill Criteria Checklist

A kill-criteria checklist that sets the stop conditions for a migration or program before it starts.

Infrastructure
Moving workloads from on premises to the cloud