Twenty years of the Java Virtual Machine Tool Interface: A systematic literature review

Context: The Java Virtual Machine Tool Interface (JVMTI) has been an essential part of the Java ecosystem for over two decades, enabling profiling, debugging, and runtime monitoring of Java applications. However, a substantial body of JVMTI-related research engages with JVMTI indirectly through derived frameworks such as DiSL and ShadowVM, rendering it partially invisible to keyword-based search. Objective: To assess the state of research on JVMTI as both a directly used API and a substrate for derived frameworks, we conducted a systematic literature review following the guidelines of Kitchenham and Charters. Methods: We adopted a two-stream search strategy combining automated keyword-based search across five academic databases (Elsevier Scopus, dblp, IEEE Xplore, ACM Digital Library, SpringerLink) with forward snowballing seeded from the database-stream inclusions. After deduplication, eligibility assessment, conflict resolution, and four iterations of snowballing until saturation, 40 studies were selected: 24 from the database search and 16 from snowballing. Results: We identify seven primary usage scenarios: debugging, performance profiling, security enforcement, fault injection, memory management, bytecode instrumentation, and JVMTI-based instrumentation frameworks as a cross-cutting category. ClassFileLoadHook emerges as the most referenced event, serving as the transport layer for load-time bytecode rewriting in both directly-coded agents and framework-based tools. Sixteen studies (40% of the corpus) use JVMTI indirectly through derived frameworks, a cluster that is inaccessible to keyword-based search and that reveals a coordinated post-2018 research program on parallel and concurrent JVM workload profiling. Conclusion: JVMTI’s scientific impact extends well beyond direct API usage, and a dual-stream search design is methodologically necessary for SLRs on software-infrastructure topics. We identify sampling-based event processing, dynamic event filtering, JIT-compilation-aware analysis, and bridging the academic-industrial gap as open avenues for future research.