DMS, our deterministic toolkit that combines symbolic AI and enhanced compiler technology, in production since 1995, migrates ADABAS data to a relational target and translates the Natural, COBOL, and PL/I that talk to it. Business logic comes out functionally equivalent to the original. Gen AI handles the adjacent work, like business rule extraction and test generation.
Trusted By
Natural/Adabas was designed in 1971 as an inverted-list system. It stores records differently than a relational database, and applications use access patterns that don’t translate one-to-one to SQL.
A single ADABAS record can hold many values for one field. In SQL: a one-to-many relationship needing its own child table. We generate the table, the foreign key, and the access code that reproduces the original behavior.
Groups of fields that repeat together inside a record. These need their own table with parent reference and occurrence counter. Naive migrations flatten PE groups (data loss) or dump them as JSON (query nightmares).
ADABAS ISNs are physical record locators, not stable primary keys. We translate ISN access into surrogate keys with proper indexing.
Composite index keys assembled from byte ranges of one or more fields, so the key has no column of its own in the record. We rebuild each one per target: an expression index on PostgreSQL, Oracle and Db2, and a persisted computed column with an index on SQL Server, which has no expression-index equivalent.
Most Natural and ADABAS shops have COBOL or PL/I that issues ADABAS commands directly through the control block and the five buffers (L1 to L6 reads, L9 histogram, S1, S2, S4 finds, A1 update, N1 and N2 store, E1 delete, ET and BT). We rewrite those sites to JDBC with the same locking and transaction boundaries.
Any credible migration plan has to account for all five. The data model rebuild is the part that gets underestimated.
Normalized relational schema, idiomatic Java, repeatable data load. How major ADABAS structures map:
Compiles. Runs. Produces the same outputs against the same inputs at the same precision. A Java developer can read it without learning Natural/Adabas.
Two layers. Translation runs on DMS. Discovery and review run on Gen AI.
The split matters because financial and government Natural/Adabas workloads can’t tolerate silent semantic drift. Gen AI guessing what an L3 call means is a defect.
ADABAS for Mainframe on z/OS and z/VSE (versions 7 and 8)
ADABAS on BS2000
ADABAS for Linux, Unix, and Windows
ADABAS C, the classic inverted-list engine, and ADABAS D, the separate relational product that later became SAP DB and MaxDB
ADABAS SQL Gateway (CONNX-based) sources
MU fields and PE groups including MU-within-PE nesting, elementary and sub-descriptors, superdescriptors, hyperdescriptors, phonetic descriptors, null suppression, and fields defined with the fixed-storage FI option
ADASEL, ADAREP, ADAULD, ADACMP for data extraction
FIND, READ, HISTOGRAM, STORE, UPDATE, DELETE with their ADABAS underpinnings
ISN-based access, L1/L2/L3 reads, S1/S2/S4 finds, ET/BT transactions
DBID, FNR, file definitions from Predict or ADAREP
Direct CALL ADABAS from COBOL, PL/I, and Assembler, across the control block, format buffer, record buffer, search buffer and value buffer: OP and CL session control, L1 to L6 record reads, L9 histogram, S1, S2, S4, S8 and S9 finds and sorts, A1 update, N1 and N2 store, E1 delete, ET and BT transaction control, and RC and RE command-ID release
Natural is covered here alongside the data migration. For the language on its own, see our Natural to Java migration page
ADABAS rarely sits alone. Most of what surrounds it has to move at the same time:
translates to Java with database calls rewritten to JDBC or JPA.
get CALL ADABAS sites rewritten to SQL blocks. The rest follows our COBOL or PL/I to Java rules.
move to Spring Boot services or REST endpoints. IBM acquired EntireX with webMethods on 1 July 2024.
translate to web UI scaffolding your team finishes. ApplinX moved to IBM in the same transaction.
becomes the schema source. JPA entities generated from the same metadata.
translate to your chosen orchestrator. Spring Batch, Airflow, and Kubernetes CronJobs are common landings.
Run ADABAS metadata and the surrounding source through DMS. Complexity report, MU/PE density, custom calls flagged.
Deliverable: Complexity report + fixed-price quote within 10 business days. The assessment is free.
Migrate 5 to 10 ADABAS files plus the Natural and COBOL that touches them. Compare against the original for functional equivalence.
Deliverable: Working Java + relational schema + equivalence results. Measured against a 95% functional equivalence threshold.
Migrate the rest of the codebase and data using rules calibrated during the pilot.
Deliverable: Rolling module deliveries every 4 to 6 weeks. Data load scripts run repeatedly for refresh.
New Java + relational stack runs alongside Natural and ADABAS, processing the same inputs and comparing outputs.
Deliverable: Zero-diff signoff. Cutover plan. ADABAS data archived.
General questions
Natural and ADABAS MU fields become child tables with a foreign key to the parent and an occurrence index. PE groups also become child tables, one per group, with the same parent reference plus a sequence number. The generated Java repository methods reproduce the original access pattern so application code behaves the same against the relational schema as it did against Natural and ADABAS.
Natural and ADABAS to Java migration generates schemas for PostgreSQL, SQL Server, Oracle, Db2, Amazon Aurora, and Azure SQL. Target choice depends on your team's existing operations stack and cloud preference. The toolkit is target-aware: it generates appropriate types, functional indexes, and DDL for the database you pick during assessment.
A typical migration covering 500 ADABAS files and 1 million lines of surrounding Natural and COBOL runs 6 to 15 months end-to-end. Smaller installations (100 to 200 files) finish much faster.
Migration is scoped per project, not per line of code. Cost depends on codebase size, dialect coverage, how much surrounding code has to move with it, and the target architecture. The free assessment produces a fixed-price quote for the whole engagement, locked for 90 days, so the number is settled before any translation work starts.
Natural and ADABAS migration will deliver 100% functional equivalence in final delivery.
ADABAS is fast on its specific access patterns. A naive migration will be slower. Our process includes performance profiling during pilot and full migration. Queries that miss the original benchmark get rewritten before cutover, and indexing gets tuned per-table. Most production cutovers reach 1:1 or better throughput on the relational target.
Yes. Natural and ADABAS to Java migration covers COBOL and PL/I programs that issue direct ADABAS commands (L1, L2, L3, S1, S2, S4, ET, BT, A1, N1, N2). The CALL ADABAS sites get rewritten to JDBC blocks with equivalent transaction and locking semantics. The rest of the COBOL or PL/I follows our standard rules for those languages.
The free assessment delivers six artifacts in 10 business days: a complexity report (file count, MU and PE density, embedded call inventory), a relational data model preview, a sample migration of 2 to 3 ADABAS files plus the code that reads them, a risk register, a phased project plan, and a fixed-price quote locked for 90 days.
COBOL headcount is falling 8% a year. Contractor rates are climbing 12 to 18% annually. Start with a free assessment. Decide afterward.
.png)
info@semanticdesigns.com
