Corpus-Guided Compression of Modern Native Executables

Motivation

Lossless compression can reduce the storage, transfer, and deployment cost of executable programs, while faster decompression can improve startup and responsiveness. For executables, compression must preserve the program bit-exactly. This matters for modern ahead-of-time (AOT) native artifacts, which are increasingly used in startup-sensitive workloads such as microservices, serverless functions, and cloud applications. Native executables produced by conventional toolchains, GraalVM Native Image, and .NET NativeAOT differ in layout, metadata, embedded runtime components, and build-time transformations, so their compression behavior cannot be assumed to be uniform.

Approach

This ongoing work studies compression of binary native executables from the perspective of corpus characteristics. We compare compressors and examine how measurable file-level properties relate to compression performance (compression ratio, compression speed, and decompression speed). The analysis is organized around executable families and build configurations, with the goal of understanding when conclusions from generic benchmark corpora transfer to modern native executables and when a specialized executable corpus is needed.

To support this analysis, we are developing NEXUS (Native EXecUtable Set), a corpus of binary programs spanning common native executables, GraalVM native images, and .NET NativeAOT images. Based on corpus characteristics and measured compression performance, we aim to provide a methodology and tools for selecting a suitable compression strategy for a given executable family and deployment objective (e.g., minimizing artifact size, maximizing decompression speed, or balancing size reduction against processing cost).

Status

This is ongoing work. Current efforts focus on corpus construction, characterization, and evaluation across compressors, executable families, and build configurations.

Acknowledgements

This work is developed as part of the research collaboration between the Faculty of Mathematics, University of Belgrade, and Oracle.