A Review of Binary Optimization in COBOL

Legacy software systems, specifically those that use COBOL, remain foundational in fields like banking, insurance, and government infrastructure. As hardware architectures evolve, optimizing legacy binaries while maintaining compatibility becomes increasingly important to harness the full potential of modern systems. However, direct refactoring of source code is often infeasible due to risk, cost, or unavailability of the original codebase. Binary optimization offers a compelling alternative that enables performance improvements at the binary level without touching the source. COBOL, one of the oldest high-level programming languages, continues to run critical workloads in sectors such as finance, government, and infrastructure. Despite its reliability, COBOL systems face growing challenges due to aging codebases, limited source code availability, and the increasing gap between legacy software and modern hardware capabilities. Recompilation is often infeasible, prompting the need for safe and practical binary-level optimization techniques. This paper presents a design-focused review of binary optimization in the context of COBOL, beginning with an overview of the language’s execution model and architectural constraints. We discuss IBM’s Automatic Binary Optimizer (ABO), a production-grade tool designed to enhance the performance of COBOL binaries on modern IBM Z systems without requiring source code. ABO’s use of “smart binaries” and instruction-level metadata enables advanced validation workflows. We also survey key research contributions such as optimize-time validation (Koju et al.), hardware idiom recognition, adaptive runtime frameworks like COBRA, and formal verification tools like Alive2.