Architecture of an expert system for analysing memory snapshoots

Abstract

This paper proposes a rule-based expert system architecture designed to analyse software application memory snapshots in order to identify and eliminate inefficient memory usage. The system uses production rules based on knowledge about object types and their properties (in particular, immutability), allocation statistics, and object distribution in memory. A formal model for representing a memory snapshot as a set of objects with certain attributes has been developed, and criteria for identifying duplicates of immutable information and excessive memory reservation have been introduced. Mechanisms for grouping objects by type and value to diagnose data duplication are described, as well as methods for identifying inefficient allocations, in particular, excessive data structure size relative to their useful content. The system generates recommendations in the form of instructions for code optimisation, namely the implementation of caching or object pools, redistribution of data structures, etc. The system evaluates the expected memory savings. Mathematical formalizations are provided to describe the input data, grouping operations, filtering, and object comparison.