masfenix
masfenix

Reputation: 7996

how to make a c# thread-safe random number generator

I have a loop in my code

Parallel.For(0, Cnts.MosqPopulation, i => { DoWork() });

however in the DoWork() function, there are multiple calls to a random number generator which is defined as follows:

public static class Utils
{
    public static readonly Random random = new Random();

}

It's static instance so that it is only seeded once. And I can use it throughout the code.

According to MSDN and other stackoverflow threads, this is not threadsafe. Infact, I have noticed at sometimes my code breaks and the random number generator starts generating all zeros (as per the MSDN documentation).

There are other stackoverflow threads, but are rather old and the implementation is slow. I can't afford to lose time on generating the numbers as the program is a scientific computation and is running hundreds of simulations.

I havn't worked with .net since the 2.0 days and I am not sure how the language has evolved to be able to make a fast, efficient, thread-safe RNG.

Here are the previous threads:

Is C# Random Number Generator thread safe?

Correct way to use Random in multithread application

Fast thread-safe random number generator for C#

Note: because I need a fast implementation, I can not use the RNGCryptoServiceProvider which is rather slow.

Note2: I don't have a minimal working code. I don't even know where to begin as I don't know how thread-safety works or have high level knowledge of c#. So it does seem like I am asking for a complete solution.

Upvotes: 5

Views: 3213

Answers (5)

Alexey.Petriashev
Alexey.Petriashev

Reputation: 1734

My implementation that combines best approaches from other answers (see design notes in class documentation).

/// <summary>
/// DotNet Random is not ThreadSafe so we need ThreadSafeRandom.
/// See also: https://stackoverflow.com/questions/3049467/is-c-sharp-random-number-generator-thread-safe.
/// Design notes:
/// 1. Uses own Random for each thread (thread local).
/// 2. Seed can be set in ThreadSafeRandom ctor. Note: Be careful - one seed for all threads can lead same values for several threads.
/// 3. ThreadSafeRandom implements Random class for simple usage instead ordinary Random.
/// 4. ThreadSafeRandom can be used by global static instance. Example: `int randomInt = ThreadSafeRandom.Global.Next()`.
/// </summary>
public class ThreadSafeRandom : Random
{
    /// <summary>
    /// Gets global static instance.
    /// </summary>
    public static ThreadSafeRandom Global { get; } = new ThreadSafeRandom();

    // Thread local Random is safe to use on that thread.
    private readonly ThreadLocal<Random> _threadLocalRandom;

    /// <summary>
    /// Initializes a new instance of the <see cref="ThreadSafeRandom"/> class.
    /// </summary>
    /// <param name="seed">Optional seed for <see cref="Random"/>. If not provided then random seed will be used.</param>
    public ThreadSafeRandom(int? seed = null)
    {
        _threadLocalRandom = new ThreadLocal<Random>(() => seed != null ? new Random(seed.Value) : new Random());
    }

    /// <inheritdoc />
    public override int Next() => _threadLocalRandom.Value.Next();

    /// <inheritdoc />
    public override int Next(int maxValue) => _threadLocalRandom.Value.Next(maxValue);

    /// <inheritdoc />
    public override int Next(int minValue, int maxValue) => _threadLocalRandom.Value.Next(minValue, maxValue);

    /// <inheritdoc />
    public override void NextBytes(byte[] buffer) => _threadLocalRandom.Value.NextBytes(buffer);

    /// <inheritdoc />
    public override void NextBytes(Span<byte> buffer) => _threadLocalRandom.Value.NextBytes(buffer);

    /// <inheritdoc />
    public override double NextDouble() => _threadLocalRandom.Value.NextDouble();
}

Upvotes: 3

Florian Doyon
Florian Doyon

Reputation: 4186

Using the ThreadStatic attribute and a custom getter, you will get a single Random instance per thread. If this is not acceptable, use locks.

public static class Utils
{
    [ThreadStatic]
    private static Random __random;

    public static Random Random => __random??(__random=new Random());
}

The ThreadStatic attribute does not run the initializer on each thread so you are responsible for doing so in your accessor. Also think about your seed initializer, you can use something like

new Random((int) ((1+Thread.CurrentThread.ManagedThreadId) * DateTime.UtcNow.Ticks) )

Upvotes: 8

Sir Rufo
Sir Rufo

Reputation: 19106

You can inherit from Random to build a thread safe random class

public class ThreadsafeRandom : Random
{
    private readonly object _lock = new object();

    public ThreadsafeRandom() : base() { }
    public ThreadsafeRandom( int Seed ) : base( Seed ) { }

    public override int Next()
    {
        lock ( _lock )
        {
            return base.Next();
        }
    }

    public override int Next( int maxValue )
    {
        lock ( _lock )
        {
            return base.Next( maxValue );
        }
    }

    public override int Next( int minValue, int maxValue )
    {
        lock ( _lock )
        {
            return base.Next( minValue, maxValue );
        }
    }

    public override void NextBytes( byte[ ] buffer )
    {
        lock ( _lock )
        {
            base.NextBytes( buffer );
        }
    }

    public override double NextDouble()
    {
        lock ( _lock )
        {
            return base.NextDouble();
        }
    }
}

and use an instance of that class

public static class Utils
{
    public static readonly Random random = new ThreadsafeRandom();

}

Upvotes: 0

vcsjones
vcsjones

Reputation: 141638

I would consider something like this:

private static int _tracker = 0;

private static ThreadLocal<Random> _random = new ThreadLocal<Random>(() => {
    var seed = (int)(Environment.TickCount & 0xFFFFFF00 | (byte)(Interlocked.Increment(ref _tracker) % 255));
    var random = new Random(seed);
    return random;
});

I'm not a huge fan of ThreadStatic these days. We have better tools than that using ThreadLocal. Just use _random.Value in your parallel loop and it will give you a new Random per thread.

It combines an atomically incrementing value as well as the default behavior of using Environemnt.TickCount. The incrementing value is there to solve the problem of two Random's getting the same seed. Note that this approach only allows 255 randoms to be created. If you need more, then change the size of the mask.

As you already noted, this isn't usable for secure purposes.

Upvotes: 2

ballard26
ballard26

Reputation: 37

If you have knowledge of how many threads you are running in parallel this may work:

Random rand = new Random();
var randomNums = Enumerable.Range(0, Cnts.MosqPopulation)
                           .Select(_ => rand.Next()).ToList();
Parallel.For(0, Cnts.MosqPopulation, i => 
{
    Random localRand = new Random(randomNums[i]);
    DoWork();
});

Not sure how indistinguishable the resulting distribution would be from a uniform one though.

Upvotes: 0

Related Questions