GPars: Parallelism the Right Way
0 likes1,233 views
This document introduces GPars, a Groovy/Java framework for expressing parallelism. It discusses how Moore's Law has led to ubiquitous parallelism due to more cores. GPars makes parallel programming easier by offering models like actors, dataflow, and CSP. It builds on Java concurrency utilities and works well for Groovy scripts and Java. GPars allows parallel collection operations to speed up tasks like calculating pi in parallel.
![@Grab ( 'org.codehaus.gpars:gpars:1.0-SNAPSHOT' )
import groovyx.gpars.GParsPool
void execute ( final int numberOfTasks ) {
GParsPool.withPool {
final int n = 100000000i
final double delta = 1.0d / n
final startTimeNanos = System.nanoTime ( )
final int sliceSize = n / numberOfTasks
final items = [ ]
for ( int i in 0i ..< numberOfTasks ) { items << i }
final pi = 4.0d * delta * items.collectParallel { taskId ->
final int start = 1i + taskId * sliceSize
final int end = ( taskId + 1i ) * sliceSize
double sum = 0.0d ;
for ( int i in start .. end ) {
final double x = ( i - 0.5d ) * delta
sum += 1.0d / ( 1.0d + x * x )
}
sum
}.sumParallel ( )
final elapseTime = ( System.nanoTime ( ) - startTimeNanos ) / 1e9
println ( '==== Groovy GPars GParsPool pi = ' + pi )
println ( '==== Groovy GPars GParsPool iteration count = ' + n )
println ( '==== Groovy GPars GParsPool elapse = ' + elapseTime )
println ( '==== Groovy GPars GParsPool processor count = ' + Runtime.getRuntime ( ).availableProcessors ( ) )
println ( '==== Groovy GPars GParsPool task count = ' + numberOfTasks )
}
}
Copyright © 2011 Russel Winder 10](https://image.slidesharecdn.com/gparsparallelismtherightway-111127062257-phpapp02/85/GPars-Parallelism-the-Right-Way-10-320.jpg)
![@Grab ( 'org.codehaus.gpars:gpars:1.0-SNAPSHOT' )
import groovyx.gpars.ParallelEnhancer
void execute ( final int numberOfTasks ) {
final int n = 100000000i
final double delta = 1.0d / n
final startTimeNanos = System.nanoTime ( )
final int sliceSize = n / numberOfTasks
final items = [ ]
for ( int i in 0i ..< numberOfTasks ) { items << i }
ParallelEnhancer.enhanceInstance ( items )
final pi = 4.0d * delta * items.collectParallel { taskId ->
final int start = 1i + taskId * sliceSize
final int end = ( taskId + 1i ) * sliceSize
double sum = 0.0d ;
for ( int i in start .. end ) {
final double x = ( i - 0.5d ) * delta
sum += 1.0d / ( 1.0d + x * x )
}
sum
}.sumParallel ( )
final elapseTime = ( System.nanoTime ( ) - startTimeNanos ) / 1e9
println ( '==== Groovy GPars ParallelEnhancer pi = ' + pi )
println ( '==== Groovy GPars ParallelEnhancer iteration count = ' + n )
println ( '==== Groovy GPars ParallelEnhancer elapse = ' + elapseTime )
println ( '==== Groovy GPars ParallelEnhancer processor count = ' + Runtime.getRuntime
( ).availableProcessors ( ) )
println ( '==== Groovy GPars ParallelEnhancer task count = ' + numberOfTasks )
}
Copyright © 2011 Russel Winder 11](https://image.slidesharecdn.com/gparsparallelismtherightway-111127062257-phpapp02/85/GPars-Parallelism-the-Right-Way-11-320.jpg)
GPars: Parallelism the Right Way
- 1. GPars Parallelism the Right Way Dr Russel Winder It’z Interactive Ltd russel@itzinteractive.com russel@russel.org.uk @russel_winder Copyright © 2011 Russel Winder 1
- 2. Moore's Law ● More transistors. ● More cores. ● More parallelism. ● Ubiqitous parallelism. Copyright © 2011 Russel Winder 2
- 3. Caveat ● Not all applications require parallelism. ● I/O bound best handled with single threaded event loop. Copyright © 2011 Russel Winder 3
- 4. Where Parallelism Is Needed ● Operating systems offer threads as access to the cores. ● Languages offer threads as infrastructure. Copyright © 2011 Russel Winder 4
- 5. Hummm... ● Shared memory multithreading is hard. ● Locks, semaphores, monitors, … Copyright © 2011 Russel Winder 5
- 6. Models of Management ● Actors ● Dataflow ● Communicating Sequential Processes (CSP) ● Data parallelism Copyright © 2011 Russel Winder 6
- 7. GPars ● A Groovy/Java implemented framework for easy expression of parallelism. ● Offers: – Actors – Dataflow – CSP – Data parallelism Copyright © 2011 Russel Winder 7
- 8. Builds on… ● GPars builds on java.util.concurrent and all the JSR166 goodness. Copyright © 2011 Russel Winder 8
- 9. Usage ● GPars works supremely well in Groovy scripts that manage Groovy and Java classes. ● GPars works very well as a Java API. Copyright © 2011 Russel Winder 9
- 10. @Grab ( 'org.codehaus.gpars:gpars:1.0-SNAPSHOT' ) import groovyx.gpars.GParsPool void execute ( final int numberOfTasks ) { GParsPool.withPool { final int n = 100000000i final double delta = 1.0d / n final startTimeNanos = System.nanoTime ( ) final int sliceSize = n / numberOfTasks final items = [ ] for ( int i in 0i ..< numberOfTasks ) { items << i } final pi = 4.0d * delta * items.collectParallel { taskId -> final int start = 1i + taskId * sliceSize final int end = ( taskId + 1i ) * sliceSize double sum = 0.0d ; for ( int i in start .. end ) { final double x = ( i - 0.5d ) * delta sum += 1.0d / ( 1.0d + x * x ) } sum }.sumParallel ( ) final elapseTime = ( System.nanoTime ( ) - startTimeNanos ) / 1e9 println ( '==== Groovy GPars GParsPool pi = ' + pi ) println ( '==== Groovy GPars GParsPool iteration count = ' + n ) println ( '==== Groovy GPars GParsPool elapse = ' + elapseTime ) println ( '==== Groovy GPars GParsPool processor count = ' + Runtime.getRuntime ( ).availableProcessors ( ) ) println ( '==== Groovy GPars GParsPool task count = ' + numberOfTasks ) } } Copyright © 2011 Russel Winder 10
- 11. @Grab ( 'org.codehaus.gpars:gpars:1.0-SNAPSHOT' ) import groovyx.gpars.ParallelEnhancer void execute ( final int numberOfTasks ) { final int n = 100000000i final double delta = 1.0d / n final startTimeNanos = System.nanoTime ( ) final int sliceSize = n / numberOfTasks final items = [ ] for ( int i in 0i ..< numberOfTasks ) { items << i } ParallelEnhancer.enhanceInstance ( items ) final pi = 4.0d * delta * items.collectParallel { taskId -> final int start = 1i + taskId * sliceSize final int end = ( taskId + 1i ) * sliceSize double sum = 0.0d ; for ( int i in start .. end ) { final double x = ( i - 0.5d ) * delta sum += 1.0d / ( 1.0d + x * x ) } sum }.sumParallel ( ) final elapseTime = ( System.nanoTime ( ) - startTimeNanos ) / 1e9 println ( '==== Groovy GPars ParallelEnhancer pi = ' + pi ) println ( '==== Groovy GPars ParallelEnhancer iteration count = ' + n ) println ( '==== Groovy GPars ParallelEnhancer elapse = ' + elapseTime ) println ( '==== Groovy GPars ParallelEnhancer processor count = ' + Runtime.getRuntime ( ).availableProcessors ( ) ) println ( '==== Groovy GPars ParallelEnhancer task count = ' + numberOfTasks ) } Copyright © 2011 Russel Winder 11
- 12. GPars Check it out, you know you want to. http://gpars.codehaus.org/ Copyright © 2011 Russel Winder 12
- 13. Mandatory Book Advert Python for Rookies Sarah Mount, James Shuttleworth and Russel Winder Thomson Learning Now called Cengage Learning. Developing Java Software Third Edition Russel Winder and Graham Roberts Wiley Buy these books! Copyright © 2010 Russel Winder 13
- 14. GPars Parallelism the Right Way Dr Russel Winder It’z Interactive Ltd russel@itzinteractive.com russel@russel.org.uk @russel_winder Copyright © 2011 Russel Winder 14