Fine-Tuning And Performance Analysis Of Embedded Java Environment

With the popularity of digital products such as mobile phones and PDAs, the imbeddedJava platform is developing rapidly and has earned the recognition from its users as thebasis of cross-platform software. But currently, the technology of embedded Java is still atits metaphase, plus the restriction of hardware resources on embedded platform due to thestagnation of optimizing the embedded Java environment, far from the development speed ofthat of standard version and enterprise version. Recent promotion of hardware performancein embedded systems outgrows the embedded java environment developped in early years inthat they can no longer make full use of the underlying hardware. At the same time, largerstorage and faster processor clock make room for further optimization of embedded Javaenvironment. Work in optimization of embedded Java environment not only make sense intheoretical research in program optimization but is also of value in real application.This paper begins at quantitative analysis of Java program execution, find the bottleneckin performance of interpretation-based Java Virtual Machine, and finally adopts the techniqueof dynamic compilation base on hot spot detection as a means of its optimization. Withthe conclusion of the analysis, average execution count per instruction is selected as thequantitative standards for"hot spot"of programs, the aproach of dynamic compilation istaken to translate the byte-codes of Java Virtual Machine into binary representation whichcould be executed directlly on the native platform, and a full set of solutions is both designedand implemented for optimizing embedded Java environment on ARM architecture.To be more specific, this paper makes the major contributions described below:1:Analyzing quantitatively the bottleneck for interpretational Java Virtual Machine'sperformance, and working out the critical heuristics for improving its performance, whichis to use average execution count per instruction as the measure for computational intensityto detect program hotspots and to dynamically translate Java instructions into native ones atruntime.