C# Thread safe fast(est) counter

Question

What is the way to obtain a thread safe counter in C# with best possible performance?

This is as simple as it gets:

public static long GetNextValue()
{
    long result;
    lock (LOCK)
    {
        result = COUNTER++;
    }
    return result;
}

But are there faster alternatives?

Embracery

Answer 1

This would be simpler:

return Interlocked.Increment(ref COUNTER);

MSDN Interlocked.Increment

Hampshire

This is fast, but still not fast enough in extreme performance scenarios. Nevertheless it's not like we have better alternatives. – Naamann

Answer 2

As recommended by others, the Interlocked.Increment will have better performance than lock(). Just take a look at the IL and Assembly where you will see that Increment turns into a "bus lock" statement and its variable is directly incremented (x86) or "added" to (x64).

This "bus lock" statement locks the bus to prevent another CPU from accessing the bus while the calling CPU does its operation. Now, take a look at the C# lock() statement's IL. Here you will see calls to Monitor in order to begin or end a section.

In other words, .Net lock() statement is doing a lot more than the .Net Interlocked.Increment.

So, if all you want to do is increment a variable, Interlock.Increment will be faster. Review all of the Interlocked methods to see the various atomic operations available and to find those that suit your needs. Use lock() when you want to do more complex things like multiple inter-related increments/decrements, or to serialize access to resources that are more complex than integers.

Ankerite

-1 for implementation details. It's true that locking is way slower than an atomic op, but this has nothing to do with the IL. Those function calls would be way faster than an atomic op if not for their semantics, which is not inherently required of the IL. – Aldaaldan

Answer 3

I suggest you use .NET's built in interlock increment in the System.Threading library.

The following code will increment a long variable by reference and is completely thread safe:

Interlocked.Increment(ref myNum);

Source: http://msdn.microsoft.com/en-us/library/dd78zt0c.aspx

Sancho

Answer 4

Try with Interlocked.Increment

Keele

Answer 5

As already mentioned use Interlocked.Increment

Code example from MS:

The following example determines how many random numbers that range from 0 to 1,000 are required to generate 1,000 random numbers with a midpoint value. To keep track of the number of midpoint values, a variable, midpointCount, is set equal to 0 and incremented each time the random number generator returns a midpoint value until it reaches 10,000. Because three threads generate the random numbers, the Increment(Int32) method is called to ensure that multiple threads don't update midpointCount concurrently. Note that a lock is also used to protect the random number generator, and that a CountdownEvent object is used to ensure that the Main method doesn't finish execution before the three threads.

using System;
using System.Threading;

public class Example
{
   const int LOWERBOUND = 0;
   const int UPPERBOUND = 1001;

   static Object lockObj = new Object();
   static Random rnd = new Random();
   static CountdownEvent cte;

   static int totalCount = 0;
   static int totalMidpoint = 0;
   static int midpointCount = 0;

   public static void Main()
   {
      cte = new CountdownEvent(1);
      // Start three threads. 
      for (int ctr = 0; ctr <= 2; ctr++) {
         cte.AddCount();
         Thread th = new Thread(GenerateNumbers);
         th.Name = "Thread" + ctr.ToString();
         th.Start();
      }
      cte.Signal();
      cte.Wait();
      Console.WriteLine();
      Console.WriteLine("Total midpoint values:  {0,10:N0} ({1:P3})",
                        totalMidpoint, totalMidpoint/((double)totalCount));
      Console.WriteLine("Total number of values: {0,10:N0}", 
                        totalCount);                  
   }

   private static void GenerateNumbers()
   {
      int midpoint = (UPPERBOUND - LOWERBOUND) / 2;
      int value = 0;
      int total = 0;
      int midpt = 0;

      do {
         lock (lockObj) {
            value = rnd.Next(LOWERBOUND, UPPERBOUND);
         }
         if (value == midpoint) { 
            Interlocked.Increment(ref midpointCount);
            midpt++;
         }
         total++;    
      } while (midpointCount < 10000);

      Interlocked.Add(ref totalCount, total);
      Interlocked.Add(ref totalMidpoint, midpt);

      string s = String.Format("Thread {0}:\n", Thread.CurrentThread.Name) +
                 String.Format("   Random Numbers: {0:N0}\n", total) + 
                 String.Format("   Midpoint values: {0:N0} ({1:P3})", midpt, 
                               ((double) midpt)/total);
      Console.WriteLine(s);
      cte.Signal();
   }
}
// The example displays output like the following:
//       Thread Thread2:
//          Random Numbers: 2,776,674
//          Midpoint values: 2,773 (0.100 %)
//       Thread Thread1:
//          Random Numbers: 4,876,100
//          Midpoint values: 4,873 (0.100 %)
//       Thread Thread0:
//          Random Numbers: 2,312,310
//          Midpoint values: 2,354 (0.102 %)
//       
//       Total midpoint values:      10,000 (0.100 %)
//       Total number of values:  9,965,084

The following example is similar to the previous one, except that it uses the Task class instead of a thread procedure to generate 50,000 random midpoint integers. In this example, a lambda expression replaces the GenerateNumbers thread procedure, and the call to the Task.WaitAll method eliminates the need for the CountdownEvent object.

using System;
using System.Collections.Generic;
using System.Threading;
using System.Threading.Tasks;

public class Example
{
   const int LOWERBOUND = 0;
   const int UPPERBOUND = 1001;

   static Object lockObj = new Object();
   static Random rnd = new Random();

   static int totalCount = 0;
   static int totalMidpoint = 0;
   static int midpointCount = 0;

   public static void Main()
   {
      List<Task> tasks = new List<Task>();
      // Start three tasks. 
      for (int ctr = 0; ctr <= 2; ctr++) 
         tasks.Add(Task.Run( () => { int midpoint = (UPPERBOUND - LOWERBOUND) / 2;
                                     int value = 0;
                                     int total = 0;
                                     int midpt = 0;

                                     do {
                                        lock (lockObj) {
                                           value = rnd.Next(LOWERBOUND, UPPERBOUND);
                                        }
                                        if (value == midpoint) { 
                                           Interlocked.Increment(ref midpointCount);
                                           midpt++;
                                        }
                                        total++;    
                                     } while (midpointCount < 50000);

                                     Interlocked.Add(ref totalCount, total);
                                     Interlocked.Add(ref totalMidpoint, midpt);

                                     string s = String.Format("Task {0}:\n", Task.CurrentId) +
                                                String.Format("   Random Numbers: {0:N0}\n", total) + 
                                                String.Format("   Midpoint values: {0:N0} ({1:P3})", midpt, 
                                                              ((double) midpt)/total);
                                     Console.WriteLine(s); } ));

      Task.WaitAll(tasks.ToArray());
      Console.WriteLine();
      Console.WriteLine("Total midpoint values:  {0,10:N0} ({1:P3})",
                        totalMidpoint, totalMidpoint/((double)totalCount));
      Console.WriteLine("Total number of values: {0,10:N0}", 
                        totalCount);                  
   }
}
// The example displays output like the following:
//       Task 3:
//          Random Numbers: 10,855,250
//          Midpoint values: 10,823 (0.100 %)
//       Task 1:
//          Random Numbers: 15,243,703
//          Midpoint values: 15,110 (0.099 %)
//       Task 2:
//          Random Numbers: 24,107,425
//          Midpoint values: 24,067 (0.100 %)
//       
//       Total midpoint values:      50,000 (0.100 %)
//       Total number of values: 50,206,378

https://learn.microsoft.com/en-us/dotnet/api/system.threading.interlocked.increment?view=netcore-3.0

Vibrate