CSE 598c

1
CSE 598c Virtual Machines Survey
ProposalImproving Performance for the
JVMSandra Rueda
2
Motivation

• Java Programs are widely used
• Performance is a big issue
• Execution time for Java Programs may be improved
• What options are related to the JVM?

3
JVM - Architecture Overview
runtime data areas
Execution engine
Class loader subsystem
methods
heap
Native method interface
registers

4
JVM - Execution Techniques
Java Bytecode
Java Bytecode
Interpretation
Compilation
JVM Platform Specific
JVM Platform Specific
Execution
Execution
Platform (Hardware)
Platform (Hardware)

• Interpretation and compilation have the usual
  advantages and trade-offs flexibility vs.
  performance.

5
JVM - Execution Techniques (2)
Java Bytecodes
Java Bytecodes
JIT Compilers
Interpretation
JVM Platform Specific
JVM Platform Specific
Execution
Java Processor (Hardware)
Platform (Hardware)

• JIT compilers also exhibit the traditional
  advantages and trade-offs.
• The Java Processor represents a different
  approach.

6
Java Proposed Optimizations (1)

• Bytecode optimization. Code optimization is a
  technique used in multiple environments.
• Alias analysis
• Inlining
• Variable elimination
• Memory locality improvement
• Parallel and distributed execution techniques
• Multithreading It depends on application
  parallelism

7
Java Proposed Optimizations (2)

• Dynamic compilation
• Online profiling - Hotspots
• Offline profiling
• Hybrid
• Thread synchronization
• Monitor implementation
• Thin/fat locks

is it related to JVM?
8
Java Proposed Optimizations (3)

• Remote method invocation
• Object serialization
• Resource management
• Garbage collector
• Generational
• Pretenuring (long-lived object regions)
• Concurrent collector

is it related to JVM?
9
Java Proposed Optimizations (4)

• Stack based architecture
• register based architecture
• Hardware translation
• Local-variable cache
• Adapting branch target buffer

is it related to JVM?
10
Topics to include in the survey

• Dynamic Compilation Hotspots
• Thread Synchronization
• RMI
• Garbage Collector
• Register Based Architecture
• Hardware Translation

11
References Initial Version

• Barabash et al. A parallel, incremental, mostly
  concurrent garbage collector for servers. OOPSLA,
  2003.
• Blackburn et al. Pretenuring for Java. OOPSLA,
  2001.
• Czajkowski et al. Multitasking without
  comprimise a virtual machine evolution. OOPSLA,
  2001.
• Kazi et al. Techniques for obtaining high
  performance in Java programs. ACM Computing
  Surveys, 2000.
• Levanoni et al. An On-the-Fly
  Reference-Counting Garbage Collector for Java.
  ACM Transactions on Programming Languages and
  Systems, 2006.

12
References Initial Version

• Li et al. Adapting Branch-Target Buffer to
  Improve the Target Predictability of Java Code.
  ACM Transactions on Architecture and Code
  optimization, 2005.
• Oi On the Design of the Local Variable Cache in
  a Hardware Translation-Based Java Virtual
  Machine. LCTES, 2005.
• Radhakrishnan et al. Improving Java Performance
  Using Hardware Translation. ICS 2001.
• Sandya Jazzing up JVMs with off-line profile
  data does it pay? ACM SIGPLAN, 2004.
• Suganuma et al. A Dynamic Optimization
  Framework for a Java Just-in-Time Compiler.
  OOPSLA, 2001.