Facoltà di scienze informatiche

Workload characterization of JVM languages

Sarimbekov, Aibek ; Binder, Walter (Dir.)

Thèse de doctorat : Università della Svizzera italiana, 2014 ; 2014INFO003.

Being developed with a single language in mind, namely Java, the Java Virtual Machine (JVM) nowadays is targeted by numerous programming languages. Automatic memory management, Just-In-Time (JIT) compilation, and adaptive optimizations provided by the JVM make it an attractive target for different language implementations. Even though being targeted by so many languages, the JVM has been tuned... More

Add to personal list
    Summary
    Being developed with a single language in mind, namely Java, the Java Virtual Machine (JVM) nowadays is targeted by numerous programming languages. Automatic memory management, Just-In-Time (JIT) compilation, and adaptive optimizations provided by the JVM make it an attractive target for different language implementations. Even though being targeted by so many languages, the JVM has been tuned with respect to characteristics of Java programs only -- different heuristics for the garbage collector or compiler optimizations are focused more on Java programs. In this dissertation, we aim at contributing to the understanding of the workloads imposed on the JVM by both dynamically-typed and statically-typed JVM languages. We introduce a new set of dynamic metrics and an easy-to-use toolchain for collecting the latter. We apply our toolchain to applications written in six JVM languages -- Java, Scala, Clojure, Jython, JRuby, and JavaScript. We identify differences and commonalities between the examined languages and discuss their implications. Moreover, we have a close look at one of the most efficient compiler optimizations - method inlining. We present the decision tree of the HotSpot JVM's JIT compiler and analyze how well the JVM performs in inlining the workloads written in different JVM languages.