Quantification of the Abstraction Penalty

"Quantification of the Abstraction Penalty" is an empirical study by Kerem Gurkan and myself investigating the performance cost associated with object-oriented abstractions in the Java Virtual Machine. Using the Java Microbenchmark Harness (JMH), the runtime impact of binding mechanisms and inheritance depth across three distinct JVM tiers: standard JIT, inlining-disabled, and fully interpreted mode were measured using AWS cloud hosting resources for controlled run-time environments.

While the basic binding penalty remained a modest 20% under standard JIT, our research identified a sharp "performance cliff" at inheritance depth 16 and beyond. At this threshold, the JIT’s inlining budget becomes saturated, causing execution costs to spike as the compiler can no longer devirtualize the call chain. In fully interpreted environments, these active super() call chains added a linear 17 ns of latency per level, reaching an extreme 3,646% overhead at depth 128 compared to a static baseline. This work provides a quantitative framework for navigating the tradeoff between clean modular architecture and the nanosecond-level performance requirements of low-latency systems.