Java Opcode and Runtime Data Analysis

1
Java Opcode and Runtime Data Analysis

• By Conrad Benham
• Supervisor Professor Arthur Sale

2
Project Goals

• To collect data on Javas
• Opcodes,
• Runtime System and
• Create a program to automate this.
• These statistics will be used by Professor Sale
  to create an optimised Hardware Java Virtual
  Machine (JVM).

3
Why do this? - Competitive Analysis

• Presently there are few JVMs implemented in
  hardware.
• Existing implementations rarely have published
  documented to the level necessary for advanced
  research.
• Java statistics are often difficult to obtain or
  sketchy in detail.

4
Why do this? - Research Analysis

• Statistics will identify most common opcodes.
• Allows hardware developers to focus on optimising
  a specific sub-set of the opcodes.
• Findings will be published to help other research
  efforts.
• Presently there are few tools which collect data
  of this kind.

5
So how does Java work?

• Java Source Code
• Created by Java Programmer
• Java Compiler
• Produces Java Byte Codes
• Platform Independent JVM
• Runs Java Byte Codes

FOR MORE INFO...

• Go to Sun Microsystems Java web site at
  www.java.sun.com

6
What are we looking for?

• Static Data
• collected from the byte codes.
• does not change during execution time.
• will change at compile time.
• Dynamic Data
• collected at runtime.
• will change at runtime (depending on inputs to a
  program).

7
Static Data

• Includes data found in Java class files
• e.g. the iadd instruction which adds two integer
  operands on the operand stack together and pushes
  the result back on the stack.
• Branching Instructions
• The offset of instructions resulting from the
  conditional statements (i.e. if, while, switch,
  try).

8
Types of Branching Instruction

• Forward
• i.e. if, else, switch and try.
• Backward
• i.e. loops such as for and while

9
Dynamic Data

• Includes runtime analysis
• e.g. average loop iterations, processor usage,
  memory usage.
• analysing feasibility of using registers,
  efficient heap allocation/de-allocation and
  garbage collection routines.

10
Analysis Benchmarks

• There are a number of benchmark suites available
• Including the Dhrystone and Whetstone benchmarks
• Benchmark code must be
• representative of most Java applications, they
  must follow the typical programmers coding style
• able to test a JVM in terms of of how much system
  resources it uses (e.g processor usage and memory
  usage).

11
Method of Collection

• Static Data
• Source Code Compilation
• Class File Disassembly
• File Parsing
• Dynamic Data
• Not finalised
• Memory usage patterns
• Execution times
• Other performance issues

12
Demonstration
13
Conclusion

• Relatively few organisations have collected data
  on Java, the emphasis is on code optimisation.
• The main aim of this project is to create and
  collect reusable components, data and statistics
  for other research.
• Comments and questions.