본문 바로가기
  • Home

An LLM-Assisted Heterogeneous DBMS Migration Model

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(9), pp.135~142
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : July 21, 2026
  • Accepted : August 19, 2026
  • Published : September 30, 2026

Dong-Bin Lim 1,  Eun-jin Oh 1,  Kim Jae Woong 1

1국립공주대학교

Accredited

ABSTRACT

This study proposes an LLM-assisted migration model for heterogeneous DBMS environments that integrates T0-based initial loading, LLM-based SQL transformation and normalization, deterministic SQL post-processing, and CDC-based synchronization. Unlike prior studies that treat SQL dialect translation, LLM-based SQL generation, and change replication separately, the proposed model combines these functions in a single migration workflow. Source metadata and object definitions are embedded in DBMS-pair-specific prompts, and executable SQL is extracted and corrected before target application. MariaDB-to-PostgreSQL validation confirmed table loading, procedural-object transformation, CDC application, checkpoint management, and execution logging. Source and target row counts matched for four tables containing 2,980 rows; two procedures, two functions, and two triggers were applied without SQL execution errors; and all 162 CDC events were processed without failure. No manual SQL correction was required, and total processing time was 449.328 seconds. Academically, the model structures LLM-based SQL transformation as a post-processed migration workflow; practically, it reduces manual correction and improves traceability. Future work will evaluate additional DBMS pairs, LLMs, recovery behavior, and larger workloads.

Journal Copyright Policy

No CCL information provided

Citation status

* References for papers published after 2025 are currently being built.