Difference between revisions of "Talk:Parallelism vs. Concurrency"

From HaskellWiki
Jump to navigation Jump to search
m
m (More changes to examples)
 
(11 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
=== Parallelism vs concurrency: what's the difference? ===
  
=== Parallelism vs concurrency: what's the difference? ===
   
−
'''''Visible'' side effects.'''
 +
<b><i>Visible</i> side effects:</b>
   
 
* Have a look at this <del>ugly eysore</del> "prototype definition" of <code>par</code>:
  
* Have a look at this <del>ugly eysore</del> "prototype definition" of <code>par</code>:
Line 8: Line 8:
 
par :: a -> b -> b
  
par :: a -> b -> b
 
par x y = case
  
par x y = case
−
unsafeLocalState (forkIO (evalIO x >> return ()))
 +
unsafeLocalState (forkIO (evaluate x >> return ()))
 
of
  
of
 
!_ -> y
  
!_ -> y
Line 16: Line 16:
   
 
<haskell>
  
<haskell>
−
evalIO :: a -> IO a
 +
evaluate :: a -> IO a
−
forkIO :: IO () -> IO ThreadId
 +
forkIO :: IO () -> IO ThreadId
−  
 
</haskell>
  
</haskell>
   
 
Assuming:
  
Assuming:
−
* <code>x</code> is ''well-define'' (it contains no <code>unsafe...</code> calls),
 +
* <code>x</code> is <i>well-defined</i> (it contains no <code>unsafe...</code> calls),
−
* <code>x</code> is ''well-behaved'' (not throwing exceptions or causing errors);
 +
* <code>x</code> is <i>well-behaved</i> (not throwing exceptions or causing errors);
   
 
then:
  
then:
 
# <code>forkIO</code> attaches a <code>ThreadId</code> to its argument, adds it to the work-queue and returns the identifier;
  
# <code>forkIO</code> attaches a <code>ThreadId</code> to its argument, adds it to the work-queue and returns the identifier;
 
# <code>par</code> then returns <code>y</code>;
  
# <code>par</code> then returns <code>y</code>;
−
# Some time later, <code>forkIO</code>'s argument is called, causing <code>evalIO</code> to start evaluating <code>x</code>.
 +
# Some time later, <code>forkIO</code>'s argument is called, causing <code>evaluate</code> to start evaluating <code>x</code>.
   
−
If <code>y</code> is still being evaluated when the evaluation of <code>x</code> commences, then we have elementary parallelism: concurrency, but with no visible side-effects.
 +
If <code>y</code> is still being evaluated when the evaluation of <code>x</code> commences, then we have concurrency, but with no visible side-effects - it behaves like elementary parallelism.
 
|}
  
|}
   
−
* Now have a look at this <del>equally-as-ugly</del> prototype definition for <del><code>spawnIO</code></del> <code>forkIO</code>:
 +
* Now have a look at this <del>nearly-as-ugly</del> prototype definition for <del><code>spawnIO</code></del> <code>forkIO</code>:
 
:{|
  
:{|
 
|<haskell>
  
|<haskell>
 
forkIO :: IO () -> IO ThreadId
  
forkIO :: IO () -> IO ThreadId
−
forkIO act = do let t = unsafeLocalState act
 +
forkIO act = do v <- newEmptyMVar
−
case par t () of
 +
let thr = do i <- myThreadId
−
!_ -> do i' <- itsThreadId t
 +
putMVar v i
−
case i' of
 +
act
−
Just i -> return i
 +
z <- unsafeInterleaveIO thr
−
Nothing -> ioError "forkIO"
 +
par z (takeMVar v)
 
</haskell>
  
</haskell>
   
Line 48: Line 47:
   
 
<haskell>
  
<haskell>
−
itsThreadId :: a -> IO (Maybe ThreadId)
 +
myThreadId :: IO ThreadId
  +
newEmptyMVar :: IO (MVar a)
  +
putMVar :: MVar a -> a -> IO ()
  +
takeMVar :: MVar a -> IO a
  +
par :: a -> b -> b
 
</haskell>
  
</haskell>
   
  +
Assuming <code>par</code>, <code>newEmptyMVar</code>, <code>putMVar</code>
−
Assuming:
  
−
* <code>par</code> and <code>itsThreadId</code> are primitive,
 +
and <code>takeMVar</code> are primitive,
−
* <code>itsThreadId</code> would return <code>Nothing</code> if it's argument had not been previously used by <code>par</code>;
  
   
 
then:
  
then:
−
# Evaluating <code>par t ()</code> causes a new <code>ThreadId</code> to be attached to <code>t</code> by the implementation;
 +
# An unused <code>MVar</code> is obtained: <code>v</code>;
−
# <code>itsThreadId</code> retrieves <code>i'</code>, the (possible) identifier for <code>t</code>;
 +
# the parameter <code>act</code> is used to build <code>thr</code> which will store its <code>ThreadId</code> in <code>v</code>;
  +
# <code>z</code>, the future result of <code>thr</code>, is then retrieved lazily by using <code>unsafeInterleaveIO</code>;
⚫
# <code>forkIO</code> then extracts and returns the identifier.
  
  +
# <code>par</code> then presents <code>z</code> for parallel evaluation;
−  
  +
# </code>putMVar</code> waits while <code>v</code> is empty;
−
This looks very much like elementary concurrency: parallelism, but having visible side effects.
 
  ⚫
# </code>putMVar</code> then returns the contents of <code>v</code>.
  +
  +
This is parallelism, but having visible side effects - it behaves like elementary concurrency.
 
|}
  
|}
   
−
Can either of these prototypes ever go mainstream?
 +
Can either prototype definition potentially go mainstream?
−
* As shown by it's type signature, <code>par</code> is supposed to be pure: avoiding the use of <code>unsafeLocalState</code> means making it primitive;
 +
* As shown by it's type signature, <code>par</code> is meant to have no visible effects: avoiding the use of <code>unsafeLocalState</code> means making it primitive;
  +
* While the use of <code>unsafeInterleaveIO</code> may annoy some, it being one of the earlier Haskell extensions means it's widely available.
−
* Considering it's already <code>IO</code>-based, <code>forkIO</code> without <code>unsafeLocalState</code> seems more likely.
  
   
  +
For now, using a primitive <code>par</code> (and others) to define <code>forkIO</code> looks like the simplest option...but if using <code>unsafeInterleaveIO</code> <i>really does</i> annoy you, how about [[IO in action|this]]:
−
[[IO, partible-style|This]] looks interesting:
  
  +
−
<haskell>
 +
:<haskell>
−
forkIO :: (OI -> ()) -> IO -> ThreadId
  
−
forkIO act u = let !(u1:u2:u3:_) = parts u in
 +
forkIO :: (OI -> ()) -> OI -> ThreadId
−
let t = act u1 in
 +
forkIO act u = let !(u1:u2:u3:u4:u5:_) = parts u in
−
case par t () of
 +
let !v = newEmptyMVar u1
−
!_ -> case itsThreadId t u2 of
 +
let z = let !i = myThreadId u2 in
−
Just i -> i
 +
let !_ = putMVar v i u3 in
−
Nothing -> ioError "forkIO" u3
 +
let !_ = act u4 in
  +
()
  +
in par z (takeMVar v u5)
 
</haskell>
  
</haskell>
   

Latest revision as of 20:14, 2 January 2024

Parallelism vs concurrency: what's the difference?

Visible side effects:

  • Have a look at this ugly eysore "prototype definition" of par:
par     :: a -> b -> b
par x y =  case
             unsafeLocalState (forkIO (evaluate x >> return ()))
           of
             !_ -> y

where: 

      evaluate :: a -> IO a
      forkIO   :: IO () -> IO ThreadId

Assuming:

  • x is well-defined (it contains no unsafe... calls),
  • x is well-behaved (not throwing exceptions or causing errors);

then:

  1. forkIO attaches a ThreadId to its argument, adds it to the work-queue and returns the identifier;
  2. par then returns y;
  3. Some time later, forkIO's argument is called, causing evaluate to start evaluating x.

If y is still being evaluated when the evaluation of x commences, then we have concurrency, but with no visible side-effects - it behaves like elementary parallelism.

  • Now have a look at this nearly-as-ugly prototype definition for spawnIO forkIO:
forkIO     :: IO () -> IO ThreadId
forkIO act =  do v <- newEmptyMVar
                 let thr = do i <- myThreadId
                              putMVar v i
                              act
                 z <- unsafeInterleaveIO thr
                 par z (takeMVar v)

where: 

      myThreadId   :: IO ThreadId
      newEmptyMVar :: IO (MVar a)
      putMVar      :: MVar a -> a -> IO ()
      takeMVar     :: MVar a -> IO a
      par          :: a -> b -> b

Assuming par, newEmptyMVar, putMVar and takeMVar are primitive,

then:

  1. An unused MVar is obtained: v;
  2. the parameter act is used to build thr which will store its ThreadId in v;
  3. z, the future result of thr, is then retrieved lazily by using unsafeInterleaveIO;
  4. par then presents z for parallel evaluation;
  5. putMVar waits while v is empty;
  6. putMVar then returns the contents of v.

This is parallelism, but having visible side effects - it behaves like elementary concurrency.

Can either prototype definition potentially go mainstream?

  • As shown by it's type signature, par is meant to have no visible effects: avoiding the use of unsafeLocalState means making it primitive;
  • While the use of unsafeInterleaveIO may annoy some, it being one of the earlier Haskell extensions means it's widely available.

For now, using a primitive par (and others) to define forkIO looks like the simplest option...but if using unsafeInterleaveIO really does annoy you, how about this: 

forkIO       :: (OI -> ()) -> OI -> ThreadId
forkIO act u =  let !(u1:u2:u3:u4:u5:_) = parts u in
                let !v = newEmptyMVar u1
                let z  = let !i = myThreadId u2 in
                         let !_ = putMVar v i u3 in
                         let !_ = act u4 in
                         ()
                in  par z (takeMVar v u5)

-- Atravers Tue Apr 20 06:04:10 UTC 2021

Retrieved from "https://wiki.haskell.org/index.php?title=Talk:Parallelism_vs._Concurrency&oldid=66472"