SLIDE 1
Structure Overview & Motivation Design I mplementation - - PowerPoint PPT Presentation
Structure Overview & Motivation Design I mplementation - - PowerPoint PPT Presentation
I mplementation and Optimization of Thread-Local Variables for a Race-Free Java Dialect Yi Zhang , Clark Verbrugge McGill University Structure Overview & Motivation Design I mplementation Experiments Conclusion & Future Work
SLIDE 2
SLIDE 3
Overview
implement and optimize thread local access a new semantic for Java race-free version of Java
SLIDE 4
Motivation
Complexity in racy program hard to validate the optimization many optimizations are prohibited data-race free property
SLIDE 5
Motivation
data are thread local by default and use shared directives for shared data ThreadLocal class in Java API thread-local access
SLIDE 6
Design
ThreadLocal
- bjects as wrapper
Original Design
ThreadLocal objects as wrapper access data: get(), set()
SLIDE 7
Design
Original Design each thread holds a ThreadLocalMap First, get map from thread Second, < ThreadLocal as key, value>
SLIDE 8
Design
Our Design thread-local the default option use “volatile” to specify the shared data
SLIDE 9
Design
Our Design
thread-local the default option
SLIDE 10
Design
Our Design
use “volatile” to specify the shared data accesses of data
SLIDE 11
Design
Original New
semantic thread-local is not inherent need support from ThreadLocal class thread-local is inherent with in semantics data access map searching
static: table look-up based
approach
non-static: normal access
without overhead initial value
fixed initial value manually and statically inherent initial value from
parents
automatically and at run-time
SLIDE 12
Implementation
Thread Local Accesses
at the start of thread, make local copy all reachable reference objects if that field is not volatile Class.staticField this.field shared with all threads shared with parent threads we do this through deep-copying
SLIDE 13
SLIDE 14
static fields
SLIDE 15
- bjects reachable
from static fields
SLIDE 16
- bjects reachable
from parent thread
SLIDE 17
Implementation
SLIDE 18
Implementation
static fields
SLIDE 19
Implementation
- bjects reachable
from static fields
SLIDE 20
Implementation
- bjects reachable
from non-static fields
SLIDE 21
Implementation
SLIDE 22
Implementation
SLIDE 23
Implementation
SLIDE 24
Implementation
SLIDE 25
Implementation
SLIDE 26
Implementation
SLIDE 27
Implementation
table look-up based mechanism to speed up
Object Object copy for thread 1
A.localItem A.localItem
Object copy for thread 2
A.sharedItem global table local table local table
SLIDE 28
Experiments
Implementation Environment: JikesRVM 3.1.1 Micro Benchmarks: Reads and writes operations on thread-local static field
SLIDE 29
Experiments
Micro Benchmarks
SLIDE 30
Experiments
Micro Benchmarks
SLIDE 31
Experiments
gc time of new design increases faster !! We need special garbage collector adapted to our dialect
Micro Benchmarks
SLIDE 32
Experiments
Micro Benchmarks
SLIDE 33
Experiments
Non-trivial Benchmarks:
Sable Research Group: Sun Java Tutorial: Doug Lea: Java Grande Forum Benchmark Suite: SPECJVM98:
series Producer/ Consumer (P/ C) traffic, roller coaster bank mtrt
SLIDE 34
Experiments
Non-trivial Benchmarks
most benchmarks shows comparable performance traffic benchmarks runs considerably slower 289 threads-> too many
SLIDE 35
Conclusion & Future Work
Conclusion Data-Race-Free thread-local data VS shared data thread-local and Java Virtual Machine
SLIDE 36
Conclusion & Future Work
Future Work Full data-race-free language design Improve performance of current design reduce the copying overhead reduce gc time
SLIDE 37