SlideShare a Scribd company logo

Haskell vs. F# vs. Scala

•
•
P
pt114

This document compares the parallel programming support in Haskell, F#, and Scala by looking at their language features, high-level abstractions for parallelism, and experimental results from implementing the n-body problem on multi-core systems. It finds that all three languages provide good support for parallelism through features like parallel collections, tasks, actors, and strategies/skeletons. Haskell uses the par and pseq primitives as well as evaluation strategies. F# utilizes tasks from the Task Parallel Library and async workflows. Scala supports parallel collections and actors. Experimental results on implementing the n-body problem in the languages show they can all effectively utilize multiple cores.

1 of 37
Downloaded 181 times
Haskell vs. F# vs. Scala A High-level Language Features and Parallelism Support Comparison1 Prabhat Totoo, Pantazis Deligiannis, Hans-Wolfgang Loidl Denpendable Systems Group School of Mathematical and Computer Sciences Heriot-Watt University, UK FHPC’12 Copenhagen, Denmark 15 September 2012 1 URL:http://www.macs.hw.ac.uk/~dsg/gph/papers/abstracts/fhpc12.html
Goal comparison of parallel programming support in the 3 languages - performance - programmability we look at: language features and high-level abstractions for parallelism experimental results from n-body problem on multi-cores discussion about performance, programmability and pragmatic aspects Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 1 / 29
Motivation promises of functional languages for parallel programming referential transparency due to the absence of side effects high-level abstractions through HOFs, function composition skeletons encapsulating common parallel patterns ”specify what instead of how to compute something ” SPJ: ”The future of parallel is declarative ” Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 2 / 29
Recent trends in language design mainstream languages integrating functional features in their design e.g. Generics (Java), lambda expression (C#, C++), delegates (C#) improves expressiveness in the language by providing functional constructs Multi-paradigm languages - make functional programming more approachable F# - Microsoft .NET platform - functional-oriented; with imperative and OO constructs Scala - JVM - emphasize on OO but combined with powerful functional features library support for parallel patterns (skeletons) e.g. TPL Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 3 / 29
THE LANGUAGES
Summary table Table: Summary of language features Key Features Parallelism Support Haskell pure functional, lazy evaluation, par and pseq static/inferred typing Evaluation strategies F# functional, imperative, object oriented, Async Workflows strict evaluation, static/inferred typing, TPL .NET interoperability PLINQ Scala functional, imperative, object oriented, Parallel Collections strict evaluation, strong/inferred typing, Actors Java interoperability Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 4 / 29
Haskell2 pure functional language, generally functions have no side-effects lazy by default typing: static, strong, type inference advanced type system supporting ADTs, typeclasses, type polymorphism monads used to: chain computation (no default eval order) IO monad separates pure from side-effecting computations main implementation: GHC - highly-optimised compiler and RTS 2 http://haskell.org/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 5 / 29
Haskell - parallel support includes support for semi-explicit parallelism through the Glasgow parallel Haskell (GpH) extensions GpH provides a single primitive for parallelism: par 1 p a r : : a −> b −> b to annotate expression that can usefully be evaluated in parallel (potential NOT mandatory parallelism) and a second primitive to enforce sequential ordering which is needed to arrange parallel computations: pseq 1 p s e q : : a −> b −> b Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 6 / 29
Haskell - parallel support GpH e.g. 1 f = s1 + s2 −− s e q u e n t i a l 2 f = s 1 ‘ par ‘ s 2 ‘ pseq ‘ ( s 1 + s 2 ) −− p a r a l l e l Evaluation strategies abstractions built on top of the basic primitives separates coordination aspects from main computation parallel map using strategies 1 parMap s t r a t f x s = map f x s ‘ u s i n g ‘ p a r L i s t s t r a t parList Spark each of the elements of a list using the given strategy strat e.g. rdeepseq Strategy that fully evaluates its argument Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 7 / 29
Haskell - parallel support Other abstractions for parallelism: Par monad - more explicit approach - uses IVars, put and get for communication - abstract some common functions e.g. parallel map DPH - exploits data parallelism, both regular and nested-data Eden - uses process abstraction, similar to λ-abstraction - for distributed-memory but also good performance on shared-memory machines3 3 Prabhat Totoo, Hans-Wolfgang Loidl. Parallel Haskell implementations of the n-body problem. Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 8 / 29
F#4 functional-oriented language SML-like language with OO extensions implemented on top of .NET, interoperable with languages such as C# can make use of arbitrary .NET libraries from F# strict by default, but has lazy values as well advanced type system: discriminated union (=ADT), object types (for .NET interoperability) value vs. variable 4 http://research.microsoft.com/fsharp/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 9 / 29
F# - parallel support uses the .NET Parallel Extensions library - high-level constructs to write/execute parallel programs Tasks Parallel Library (TPL) - hide low-level thread creation, management, scheduling details Tasks - main construct for task parallelism Task: provides high-level abstraction compared to working directly with threads; does not return a value Task<TResult>: represents an operation that calculates a value of type TResult eventually i.e. a future Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 10 / 29
F# - parallel support Tasks Parallel Library (TPL) Parallel Class - for data parallelism - basic loop parallelisation using Parallel.For and Parallel.ForEach PLINQ - declarative model for data parallelism - uses tasks internally Async Workflows - use the async {...} keyword; doesn’t block calling thread - provide basic parallelisation - intended mainly for operations involving I/O e.g. run multiple downloads in parallel Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 11 / 29
Scala5 designed to be a “better Java”... multiparadigm language; integrating OO and functional model evaluation: strict; typing: static, strong and inferred strong interoperability with Java (targeted for JVM) still very tied with the concept of objects language expressiveness is extended by functional features 5 http://www.scala-lang.org/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 12 / 29
Scala - parallel support Parallel Collections Framework implicit (data) parallelism use par on sequential collection to invoke parallel implementation subsequent operations are parallel use seq to turn collection back to sequential 1 x s . map ( x => f x ) // s e q u e n t i a l 2 3 x s . p a r . map ( x => f x ) . s e q // p a r a l l e l built on top of Fork/Join framework thread pool implementation schedules tasks among available processors Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 13 / 29
Scala - parallel support Actors message-passing model; no shared state similar concurrency model to Erlang lightweight processes communicates by exchanging asynchronous messages messages go into mailboxes; processed using pattern matching 1 a ! msg // s e n d message 2 3 receive { // p r o c e s s m a i l b o x 4 case msg pattern 1 => a c t i o n 1 5 case msg pattern 2 => a c t i o n 2 6 case msg pattern 3 => a c t i o n 3 7 } Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 14 / 29
IMPLEMENTATION
N-body problem problem of predicting and simulating the motion of a system of N bodies that interact with each other gravitationally simulation proceeds over a number of time steps in each step calculate acceleration of each body wrt the others, then update position and velocity solving methods: all-pairs: direct body-to-body comparison, not feasible for large N Barnes-Hut algorithm: efficient approximation method, more advanced consists of 2 phases: - tree construction - force calculation (most compute-intensive) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 15 / 29
Sequential Implementation and optimisations Approach: start off with an efficient seq implementation, preserving opportunities for parallelism generic optimisations: deforestation - eliminating multiple traversal of data structures tail-call elimination compiler optimisations Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 16 / 29
Haskell Optimisations eliminates stack overflow by making functions tail recursive add strictness annotations where required use strict fields and UNPACK pragma in datatype definitions shortcut fusion e.g. foldr/build Introduce parallelism at the top-level map function Core parallel code for Haskell 1 c h u n k s i z e = ( l e n g t h b s ) ‘ quot ‘ ( n u m C a p a b i l i t i e s ∗ 4 ) 2 n e w b s = map f b s ‘ u s i n g ‘ p a r L i s t C h u n k c h u n k s i z e rdeepseq Parallel tuning chunking to control granularity, not too many small tasks; not too few large tasks Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 17 / 29
F# (1) translate Haskell code into F# keeping the code functional, so mostly syntax changes some general optimisations apply + inlining 1 ( ∗ TPL ∗ ) 1 ( ∗ Async W o r k f l o w s ∗ ) 2 2 l e t pmap async f xs = 3 (∗ E x p l i c i t t a s k s c r e a t i o n . ∗) 3 s e q { f o r x i n x s −> a s y n c { 4 l e t p m a p t p l t a s k s f ( x s : l i s t < >) return f x } } = 4 |> Async . P a r a l l e l 5 L i s t . map ( f u n x −> 5 |> Async . R u n S y n c h r o n o u s l y 6 Task< >. F a c t o r y . S t a r t N e w ( f u n 6 |> Seq . t o L i s t ( ) −> f x ) . R e s u l t 7 ) xs 1 ( ∗ PLINQ − u s e s TPL i n t e r n a l l y ∗ ) 2 l e t p m a p p l i n q f ( x s : l i s t < >) = 3 x s . A s P a r a l l e l ( ) . S e l e c t ( f u n x −> f x ) |> Seq . t o L i s t Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 18 / 29
F# (2) imperative style 1 (∗ P a r a l l e l . For ∗) 2 3 l e t p m a p t p l p a r f o r f ( x s : a r r a y < >) = 4 l e t new xs = Array . z e r o C r e a t e xs . Length 5 P a r a l l e l . F o r ( 0 , x s . Length , ( f u n i −> 6 n e w x s . [ i ] <− f ( x s . [ i ] ) ) ) |> i g n o r e 7 new xs Parallel tuning maximum degree of parallelism - specifies max number of concurrently executing tasks chunking/partitioning - to control the granularity of tasks Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 19 / 29
Scala translate Haskell and F# implementations into Scala keeping the code as much functional as possible optimisations - unnecessary object initialisations removal 1 // P a r a l l e l C o l l e c t i o n s . 2 b o d i e s . p a r . map ( ( b : Body ) => new Body ( b . mass , u p d a t e P o s ( b ) , updateVel (b) ) ) . seq 1 // P a r a l l e l map u s i n g F u t u r e s . 2 d e f pmap [ T ] ( f : T => T) ( x s : L i s t [ T ] ) : L i s t [ T ] = { 3 v a l t a s k s = x s . map ( ( x : T) => F u t u r e s . f u t u r e { f ( x ) } ) 4 t a s k s . map ( f u t u r e => f u t u r e . a p p l y ( ) ) 5 } Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 20 / 29
RESULTS
Experimental setup Platforms and language implementations: Linux (64-bit) Windows (32-bit) 2.33GHz (8 cores) 2.80GHz (4 cores with HT) Haskell GHC 7.4.1 GHC 7.4.1 F# F# 2.0 / Mono 2.11.1 F# 2.0 / .NET 4.0 Scala Scala 2.10 / JVM 1.7 Scala 2.10 / JVM 1.7 Input size: 80,000 bodies, 1 iteration Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 21 / 29
Sequential Runtimes Haskell (Linux) 479.94s F# (Win) 28.43s Scala (Linux) 55.44s Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
Sequential Runtimes Haskell (Linux) 479.94s F# (Win) 28.43s Scala (Linux) 55.44s before Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
Sequential Runtimes Haskell (Linux) 479.94s 25.28 F# (Win) 28.43s 21.12 Scala (Linux) 55.44s 39.04 before after Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
Sequential Runtimes Haskell (Linux) 479.94s 25.28 F# (Win) 28.43s 21.12 Scala (Linux) 55.44s 39.04 before after Linux Windows Haskell 25.28 17.64 F# 118.12 21.12 Scala 39.04 66.65 Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
Parallel Runtimes Haskell (EvalStrat) F# (PLINQ) Scala (Actors) seq 25.28 118.12 39.04 1 26.38 196.14 40.01 2 14.48 120.78 22.34 4 7.41 80.91 14.88 8 4.50 70.67 13.26 Table: Linux (8 cores) Haskell (EvalStrat) F# (PLINQ) Scala (Actors) seq 17.64 21.12 66.65 1 18.05 21.39 67.24 2 9.41 17.32 58.66 4 6.80 10.56 33.84 8 (HT) 4.77 8.64 25.28 Table: Windows (4 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 23 / 29
Speedups Linux (8 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 24 / 29
Speedups Windows (4 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 25 / 29
Observations Performance Haskell outperforms F# and Scala on both platforms - has been possible after many optimisations poor performance - F#/Mono (vs. F#/.NET runtime) poor performance - Scala on Windows Programmability Haskell - implication of laziness - hard to estimate how much work is involved Scala - verbose, unlike other functional languages in general, initial parallelism easy to specify chunking required to work for Haskell, but PLINQ and ParColl are more implicit F# and Scala retain much control in the implementation - not easy to tune parallel program Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 26 / 29
Observations Pragmatics Haskell - good tool support e.g. for space and time profiling - threadscope: visualisation tool to see work distribution F# - Profiling and Analysis tools available in VS2011 Beta Pro - Concurrency Visualizer Scala - benefits from free tools available for JVM Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 27 / 29
Conclusions employ skeleton-based, semi-explicit and explicit approaches to parallelism (near) best runtimes using highest-level of abstraction start with an optimised sequential program but use impure features only after parallelisation Haskell provides least intrusive mechanism for parallelisation F# provides the preferred mechanism for data-parallelism with the SQL-like PLINQ extra programming effort using actor-based code in Scala not justified Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 28 / 29
Future Work: Eden on Clusters6 Eden Iteration Skeletons: Naive NBody with 15000 bodies, 10 iteration 512 256 128 64 Time (s) 32 16 8 localLoop/allToAllIter loopControl/allToAllRD 4 linear speedup 1 2 4 8 16 32 64 128 Processors 6 Mischa Dieterle, Thomas Horstmeyer, Jost Berthold and Rita Loogen: Iterating Skeletons - Structured Parallelism by Composition, IFL’12 Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 29 / 29
Thank you for listening! Full paper and sources: http://www.macs.hw.ac.uk/~dsg/gph/papers/abstracts/ fhpc12.html Email: {pt114, pd85, H.W.Loidl}@hw.ac.uk

More Related Content

Haskell vs. F# vs. Scala

  • 1. Haskell vs. F# vs. Scala A High-level Language Features and Parallelism Support Comparison1 Prabhat Totoo, Pantazis Deligiannis, Hans-Wolfgang Loidl Denpendable Systems Group School of Mathematical and Computer Sciences Heriot-Watt University, UK FHPC’12 Copenhagen, Denmark 15 September 2012 1 URL:http://www.macs.hw.ac.uk/~dsg/gph/papers/abstracts/fhpc12.html
  • 2. Goal comparison of parallel programming support in the 3 languages - performance - programmability we look at: language features and high-level abstractions for parallelism experimental results from n-body problem on multi-cores discussion about performance, programmability and pragmatic aspects Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 1 / 29
  • 3. Motivation promises of functional languages for parallel programming referential transparency due to the absence of side effects high-level abstractions through HOFs, function composition skeletons encapsulating common parallel patterns ”specify what instead of how to compute something ” SPJ: ”The future of parallel is declarative ” Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 2 / 29
  • 4. Recent trends in language design mainstream languages integrating functional features in their design e.g. Generics (Java), lambda expression (C#, C++), delegates (C#) improves expressiveness in the language by providing functional constructs Multi-paradigm languages - make functional programming more approachable F# - Microsoft .NET platform - functional-oriented; with imperative and OO constructs Scala - JVM - emphasize on OO but combined with powerful functional features library support for parallel patterns (skeletons) e.g. TPL Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 3 / 29
  • 6. Summary table Table: Summary of language features Key Features Parallelism Support Haskell pure functional, lazy evaluation, par and pseq static/inferred typing Evaluation strategies F# functional, imperative, object oriented, Async Workflows strict evaluation, static/inferred typing, TPL .NET interoperability PLINQ Scala functional, imperative, object oriented, Parallel Collections strict evaluation, strong/inferred typing, Actors Java interoperability Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 4 / 29
  • 7. Haskell2 pure functional language, generally functions have no side-effects lazy by default typing: static, strong, type inference advanced type system supporting ADTs, typeclasses, type polymorphism monads used to: chain computation (no default eval order) IO monad separates pure from side-effecting computations main implementation: GHC - highly-optimised compiler and RTS 2 http://haskell.org/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 5 / 29
  • 8. Haskell - parallel support includes support for semi-explicit parallelism through the Glasgow parallel Haskell (GpH) extensions GpH provides a single primitive for parallelism: par 1 p a r : : a −> b −> b to annotate expression that can usefully be evaluated in parallel (potential NOT mandatory parallelism) and a second primitive to enforce sequential ordering which is needed to arrange parallel computations: pseq 1 p s e q : : a −> b −> b Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 6 / 29
  • 9. Haskell - parallel support GpH e.g. 1 f = s1 + s2 −− s e q u e n t i a l 2 f = s 1 ‘ par ‘ s 2 ‘ pseq ‘ ( s 1 + s 2 ) −− p a r a l l e l Evaluation strategies abstractions built on top of the basic primitives separates coordination aspects from main computation parallel map using strategies 1 parMap s t r a t f x s = map f x s ‘ u s i n g ‘ p a r L i s t s t r a t parList Spark each of the elements of a list using the given strategy strat e.g. rdeepseq Strategy that fully evaluates its argument Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 7 / 29
  • 10. Haskell - parallel support Other abstractions for parallelism: Par monad - more explicit approach - uses IVars, put and get for communication - abstract some common functions e.g. parallel map DPH - exploits data parallelism, both regular and nested-data Eden - uses process abstraction, similar to λ-abstraction - for distributed-memory but also good performance on shared-memory machines3 3 Prabhat Totoo, Hans-Wolfgang Loidl. Parallel Haskell implementations of the n-body problem. Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 8 / 29
  • 11. F#4 functional-oriented language SML-like language with OO extensions implemented on top of .NET, interoperable with languages such as C# can make use of arbitrary .NET libraries from F# strict by default, but has lazy values as well advanced type system: discriminated union (=ADT), object types (for .NET interoperability) value vs. variable 4 http://research.microsoft.com/fsharp/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 9 / 29
  • 12. F# - parallel support uses the .NET Parallel Extensions library - high-level constructs to write/execute parallel programs Tasks Parallel Library (TPL) - hide low-level thread creation, management, scheduling details Tasks - main construct for task parallelism Task: provides high-level abstraction compared to working directly with threads; does not return a value Task<TResult>: represents an operation that calculates a value of type TResult eventually i.e. a future Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 10 / 29
  • 13. F# - parallel support Tasks Parallel Library (TPL) Parallel Class - for data parallelism - basic loop parallelisation using Parallel.For and Parallel.ForEach PLINQ - declarative model for data parallelism - uses tasks internally Async Workflows - use the async {...} keyword; doesn’t block calling thread - provide basic parallelisation - intended mainly for operations involving I/O e.g. run multiple downloads in parallel Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 11 / 29
  • 14. Scala5 designed to be a “better Java”... multiparadigm language; integrating OO and functional model evaluation: strict; typing: static, strong and inferred strong interoperability with Java (targeted for JVM) still very tied with the concept of objects language expressiveness is extended by functional features 5 http://www.scala-lang.org/ Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 12 / 29
  • 15. Scala - parallel support Parallel Collections Framework implicit (data) parallelism use par on sequential collection to invoke parallel implementation subsequent operations are parallel use seq to turn collection back to sequential 1 x s . map ( x => f x ) // s e q u e n t i a l 2 3 x s . p a r . map ( x => f x ) . s e q // p a r a l l e l built on top of Fork/Join framework thread pool implementation schedules tasks among available processors Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 13 / 29
  • 16. Scala - parallel support Actors message-passing model; no shared state similar concurrency model to Erlang lightweight processes communicates by exchanging asynchronous messages messages go into mailboxes; processed using pattern matching 1 a ! msg // s e n d message 2 3 receive { // p r o c e s s m a i l b o x 4 case msg pattern 1 => a c t i o n 1 5 case msg pattern 2 => a c t i o n 2 6 case msg pattern 3 => a c t i o n 3 7 } Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 14 / 29
  • 18. N-body problem problem of predicting and simulating the motion of a system of N bodies that interact with each other gravitationally simulation proceeds over a number of time steps in each step calculate acceleration of each body wrt the others, then update position and velocity solving methods: all-pairs: direct body-to-body comparison, not feasible for large N Barnes-Hut algorithm: efficient approximation method, more advanced consists of 2 phases: - tree construction - force calculation (most compute-intensive) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 15 / 29
  • 19. Sequential Implementation and optimisations Approach: start off with an efficient seq implementation, preserving opportunities for parallelism generic optimisations: deforestation - eliminating multiple traversal of data structures tail-call elimination compiler optimisations Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 16 / 29
  • 20. Haskell Optimisations eliminates stack overflow by making functions tail recursive add strictness annotations where required use strict fields and UNPACK pragma in datatype definitions shortcut fusion e.g. foldr/build Introduce parallelism at the top-level map function Core parallel code for Haskell 1 c h u n k s i z e = ( l e n g t h b s ) ‘ quot ‘ ( n u m C a p a b i l i t i e s ∗ 4 ) 2 n e w b s = map f b s ‘ u s i n g ‘ p a r L i s t C h u n k c h u n k s i z e rdeepseq Parallel tuning chunking to control granularity, not too many small tasks; not too few large tasks Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 17 / 29
  • 21. F# (1) translate Haskell code into F# keeping the code functional, so mostly syntax changes some general optimisations apply + inlining 1 ( ∗ TPL ∗ ) 1 ( ∗ Async W o r k f l o w s ∗ ) 2 2 l e t pmap async f xs = 3 (∗ E x p l i c i t t a s k s c r e a t i o n . ∗) 3 s e q { f o r x i n x s −> a s y n c { 4 l e t p m a p t p l t a s k s f ( x s : l i s t < >) return f x } } = 4 |> Async . P a r a l l e l 5 L i s t . map ( f u n x −> 5 |> Async . R u n S y n c h r o n o u s l y 6 Task< >. F a c t o r y . S t a r t N e w ( f u n 6 |> Seq . t o L i s t ( ) −> f x ) . R e s u l t 7 ) xs 1 ( ∗ PLINQ − u s e s TPL i n t e r n a l l y ∗ ) 2 l e t p m a p p l i n q f ( x s : l i s t < >) = 3 x s . A s P a r a l l e l ( ) . S e l e c t ( f u n x −> f x ) |> Seq . t o L i s t Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 18 / 29
  • 22. F# (2) imperative style 1 (∗ P a r a l l e l . For ∗) 2 3 l e t p m a p t p l p a r f o r f ( x s : a r r a y < >) = 4 l e t new xs = Array . z e r o C r e a t e xs . Length 5 P a r a l l e l . F o r ( 0 , x s . Length , ( f u n i −> 6 n e w x s . [ i ] <− f ( x s . [ i ] ) ) ) |> i g n o r e 7 new xs Parallel tuning maximum degree of parallelism - specifies max number of concurrently executing tasks chunking/partitioning - to control the granularity of tasks Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 19 / 29
  • 23. Scala translate Haskell and F# implementations into Scala keeping the code as much functional as possible optimisations - unnecessary object initialisations removal 1 // P a r a l l e l C o l l e c t i o n s . 2 b o d i e s . p a r . map ( ( b : Body ) => new Body ( b . mass , u p d a t e P o s ( b ) , updateVel (b) ) ) . seq 1 // P a r a l l e l map u s i n g F u t u r e s . 2 d e f pmap [ T ] ( f : T => T) ( x s : L i s t [ T ] ) : L i s t [ T ] = { 3 v a l t a s k s = x s . map ( ( x : T) => F u t u r e s . f u t u r e { f ( x ) } ) 4 t a s k s . map ( f u t u r e => f u t u r e . a p p l y ( ) ) 5 } Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 20 / 29
  • 25. Experimental setup Platforms and language implementations: Linux (64-bit) Windows (32-bit) 2.33GHz (8 cores) 2.80GHz (4 cores with HT) Haskell GHC 7.4.1 GHC 7.4.1 F# F# 2.0 / Mono 2.11.1 F# 2.0 / .NET 4.0 Scala Scala 2.10 / JVM 1.7 Scala 2.10 / JVM 1.7 Input size: 80,000 bodies, 1 iteration Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 21 / 29
  • 26. Sequential Runtimes Haskell (Linux) 479.94s F# (Win) 28.43s Scala (Linux) 55.44s Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
  • 27. Sequential Runtimes Haskell (Linux) 479.94s F# (Win) 28.43s Scala (Linux) 55.44s before Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
  • 28. Sequential Runtimes Haskell (Linux) 479.94s 25.28 F# (Win) 28.43s 21.12 Scala (Linux) 55.44s 39.04 before after Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
  • 29. Sequential Runtimes Haskell (Linux) 479.94s 25.28 F# (Win) 28.43s 21.12 Scala (Linux) 55.44s 39.04 before after Linux Windows Haskell 25.28 17.64 F# 118.12 21.12 Scala 39.04 66.65 Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 22 / 29
  • 30. Parallel Runtimes Haskell (EvalStrat) F# (PLINQ) Scala (Actors) seq 25.28 118.12 39.04 1 26.38 196.14 40.01 2 14.48 120.78 22.34 4 7.41 80.91 14.88 8 4.50 70.67 13.26 Table: Linux (8 cores) Haskell (EvalStrat) F# (PLINQ) Scala (Actors) seq 17.64 21.12 66.65 1 18.05 21.39 67.24 2 9.41 17.32 58.66 4 6.80 10.56 33.84 8 (HT) 4.77 8.64 25.28 Table: Windows (4 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 23 / 29
  • 31. Speedups Linux (8 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 24 / 29
  • 32. Speedups Windows (4 cores) Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 25 / 29
  • 33. Observations Performance Haskell outperforms F# and Scala on both platforms - has been possible after many optimisations poor performance - F#/Mono (vs. F#/.NET runtime) poor performance - Scala on Windows Programmability Haskell - implication of laziness - hard to estimate how much work is involved Scala - verbose, unlike other functional languages in general, initial parallelism easy to specify chunking required to work for Haskell, but PLINQ and ParColl are more implicit F# and Scala retain much control in the implementation - not easy to tune parallel program Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 26 / 29
  • 34. Observations Pragmatics Haskell - good tool support e.g. for space and time profiling - threadscope: visualisation tool to see work distribution F# - Profiling and Analysis tools available in VS2011 Beta Pro - Concurrency Visualizer Scala - benefits from free tools available for JVM Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 27 / 29
  • 35. Conclusions employ skeleton-based, semi-explicit and explicit approaches to parallelism (near) best runtimes using highest-level of abstraction start with an optimised sequential program but use impure features only after parallelisation Haskell provides least intrusive mechanism for parallelisation F# provides the preferred mechanism for data-parallelism with the SQL-like PLINQ extra programming effort using actor-based code in Scala not justified Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 28 / 29
  • 36. Future Work: Eden on Clusters6 Eden Iteration Skeletons: Naive NBody with 15000 bodies, 10 iteration 512 256 128 64 Time (s) 32 16 8 localLoop/allToAllIter loopControl/allToAllRD 4 linear speedup 1 2 4 8 16 32 64 128 Processors 6 Mischa Dieterle, Thomas Horstmeyer, Jost Berthold and Rita Loogen: Iterating Skeletons - Structured Parallelism by Composition, IFL’12 Totoo, Deligiannis, Loidl (HWU) Haskell vs. F# vs. Scala 15-Sep-2012 29 / 29
  • 37. Thank you for listening! Full paper and sources: http://www.macs.hw.ac.uk/~dsg/gph/papers/abstracts/ fhpc12.html Email: {pt114, pd85, H.W.Loidl}@hw.ac.uk