AsParallel.ForAll vs Parallel.ForEach

Question

Is there any difference between the below code snippets. If so, what?

myList.AsParallel().ForAll(i => { /*DO SOMETHING*/ });

and

Parallel.ForEach(mylist, i => { /*DO SOMETHING*/ });

Will the main thread wait for all the child threads to complete? In a MVC application, if I'm doing parallel processing in my controller action, what happens to the child threads after the main thread completes. Will they be aborted or will they be completed even after the main thread is completed?

Solution 1

Parallel.ForEach() is intended exactly for this kind of code.

On the other hand, ForAll() is intended to be used at the end of a (possibly complex) PLINQ query.

Because of that, I think Parallel.ForEach() is the better choice here.

In both cases, the current thread will be used to perform the computations (along with some threads from the thread pool) and the method will return only after all processing has been completed.

Solution 2

Here is an explanation in MSDN:

https://docs.microsoft.com/en-us/dotnet/standard/parallel-programming/potential-pitfalls-with-plinq#prefer-forall-to-foreach-when-it-is-possible

Based on what I read, use Parallel.ForEach() if you want to ensure the list are access sequentially while using AsParallel.ForAll() does not guarantee the items in the list are accessed in order.

For MVC, all thread are request-based. The caller thread (main) is blocked until the Parallel() call is completed, that means all child threads should have completed as well.

If the caller thread is aborting, here is an explain:

http://www.albahari.com/threading/part5.aspx

PLINQ doesn't preemptively abort threads, because of the danger of doing so. Instead, upon cancellation it waits for each worker thread to finish with its current element before ending the query. This means that any external methods that the query calls will run to completion.

Solution 3

As written in Concurrency in C# Cookbook:

One difference between Parallel and PLINQ is that PLINQ assumes it can use all of the cores on the computer, while Parallel will dynamically react to changing CPU conditions.

Comments

[e_ne]

I suggest you to read: reedcopsey.com/2010/02/03/…

[Pacane]

If I were you, I would open another question for your MVC parallel processing thingy.

[YukiSakura]

Please be careful of AsParallel().ForAll() as it causes unpredictive result. For example I have a button to execute this code when clicked: myEnumerable.AsParallel().ForAll(i as string => otherDictionary.Add(i, 0)) . It will add null as a key to otherDictionary. I had to rewrote to use foreach loop. Weird.

[crokusek]

@YukiSakura, Maybe you weren't using a ConncurrentDictionary? I think we should not be afraid use use code because of a comment without a full example. It would be better to post your issue as a separate question.

[Aran Mulholland]

Why is ForAll() intended to be used at the end of a PLINQ query?

[Kabbalah]

@AranMulholland According to Reed Copsey a parallel query that is consumed by Parallel.Foreach() has to pay the cost for parallization twice. Whereas ForAll() uses existing partitions/threads from the query.

[Ben]

I guess that since Parallel.ForEach() is from the System.Threading.Tasks namespace then it is more likely related to tasks. While AsParallel().ForAll is from the System.Linq namespace and so is more likely to be better used with PLINQ.

[svick]

@Ben I don't quite understand ForAll is part of PLINQ. And I don't know what "related to tasks" means.

[svick]

AsOrdered() affects the ordering of results. And since ForAll() doesn't have any results, AsOrdered() won't have any effect here. If you're talking about ordering of processing, that makes sense only for single-threaded computations.

[thewpfguy]

Am I saying you use ForAll() with AsOrdered()? Please check.

[svick]

Sorry, my mistake. But using AsOrdered with Parallel.ForEach makes even less sense, the original non-parallel collection is already ordered, so adding .AsParallel().AsOrdered() won't have any effect. Besides, trying to order parallel computations rarely makes sense.