Neil
Neil

Reputation: 666

Parallel ForEach Giving different results in different ocasions

I have never used Parallel.ForEach but I played around with it and found this occurrence.

I run a parallel loop (code found on msdn https://msdn.microsoft.com/en-us/library/dd997393(v=vs.110).aspx did edit it with the subtotal *=2 to try understand what it is doing) with an enumerable range firstly (0,1) and then (0,1,2) then I run the second one again but after sleeping the thread for 200 milliseconds, and the result then differs

if the Thread.sleep(200)is not commented out this is the results

result 1 = 2
result 2 = 6
result 3 = 4

if the Thread.sleep(200)is commented out this is the results

result 1 = 2 
result 2 = 6
result 3 = 6

Here is the code

Stopwatch timer = new Stopwatch();
int[] nums = Enumerable.Range(0, 1).ToArray();
long total = 0;
for (int i = 0; i < 2; i++)
{
     timer.Restart();
     total = 0;
     if (i == 0) nums = Enumerable.Range(0, 1).ToArray();
     if (i == 1) nums = Enumerable.Range(0, 2).ToArray();
     Parallel.ForEach<int, long>(nums,() => 0,(j, loop, subtotal) =>
     {
       subtotal += 1;
       subtotal *= 2;
       return subtotal;
     },(finalResult) => Interlocked.Add(ref total, finalResult)); 

     Console.WriteLine("The total from Parallel.ForEach is {0:N0} and took {1}", total, timer.Elapsed);
     timer.Stop();
     //Thread.Sleep(200);
}

timer.Restart();
nums = Enumerable.Range(0, 2).ToArray();
total = 0;
Parallel.ForEach<int, long>(nums, () => 0,  (j, loop, subtotal) =>
{
    subtotal += 1;
    subtotal *= 2;
    return subtotal;
},(finalResult) => Interlocked.Add(ref total, finalResult)); 

Console.WriteLine("The total from Parallel.ForEach is {0:N0} and took {1}", total, timer.Elapsed);
timer.Stop();

I figured it has got to do with threads working over one another but this seems like a bug

Note I did have a look at Simulation gives different result with normal for loop Vs Parallel For
Why is this happening?

Upvotes: 1

Views: 1690

Answers (1)

Damien_The_Unbeliever
Damien_The_Unbeliever

Reputation: 239664

Because this code is ill-defined:

 Parallel.ForEach<int, long>(nums,() => 0,(j, loop, subtotal) =>
 {
   subtotal += 1;
   subtotal *= 2;
   return subtotal;
 },(finalResult) => Interlocked.Add(ref total, finalResult));

In that, if a single thread executes both iterations then you get the result 6. Effectively, you do:

subTotal = 0; //From init
subTotal += 1; //=1 First iteration
subTotal *= 2; //=2 First iteration
subTotal += 1; //=3 Second iteration
subTotal *= 2; //=6 Second iteration
total += subTotal; //=6 End gathering (actually interlocked)

But if two threads share the work, you get

subTotal1 = 0; //From init
subTotal2 = 0; //From init
subTotal2 += 1; //=1
subTotal1 += 1; //=1
subTotal1 *= 2; //=2
subTotal2 *= 2; //=2
total += subTotal1 //=2 End gathering 1 (interlocked)
total += subTotal2 //=4 End gathering 2 (interlocked)

Upvotes: 3

Related Questions