Using stdlib's rand() from multiple threads

Question

I have several threads which all run the same function. In each of these they generate a different random number several times. We tried to do this by putting srand(time(0)) at the start of the function, but it seems that they all get the same number.

Do we need to call srand(time(0)) only once per program, i.e at the start of main (for example), at the start of each function that is called several times, or something else?

Solution 1

srand() seeds the random number generator. You should only have to call srand(time(NULL)) once during startup.

That said, the documentation states:

The function rand() is not reentrant or thread-safe, since it uses hidden state that is modified on each call. This might just be the seed value to be used by the next call, or it might be something more elaborate. In order to get reproducible behaviour in a threaded application, this state must be made explicit. The function rand_r() is supplied with a pointer to an unsigned int, to be used as state. This is a very small amount of state, so this function will be a weak pseudo-random generator. Try drand48_r(3) instead.

The emphasized part of the above is probably the reason why all your threads get the same number.

Solution 2

As you are using C++, rather than C, you may be able to avoid the threading problems often associated with srand/rand by using c++11. This depends on using a recent compiler which supports these features. You would use a separate engine and distribution on each thread. The example acts like a dice.

#include <random>
#include <functional>

std::uniform_int_distribution<int> dice_distribution(1, 6);
std::mt19937 random_number_engine; // pseudorandom number generator
auto dice_roller = std::bind(dice_distribution, random_number_engine);
int random_roll = dice_roller();  // Generate one of the integers 1,2,3,4,5,6.

I referred to Wikipedia C++11 and Boost random when answering this question.

Solution 3

From the rand man page:

The function rand() is not reentrant or thread-safe, since it uses hidden state that is modified on each call.

So don't use it with threaded code. Use rand_r (or drand48_r if you're on linux/glibc). Seed each RNG with a different value (you could seed a first RNG in the main thread to produce random seeds for the ones in each thread).

Solution 4

If you are launching the threads all at the same time, the time sent to srand is probably the same for each thread. Since they all have the same seed, they all return the same sequence. Try using something else like a memory address from a local variable.

Solution 5

C was not designed for multithreading, so behavior of srand() with multithreading is not defined and depends on the C runtime library.

Many Unix/Linux C runtime libraries use single static state, which is not safe to access from multiple threads, so with these C runtimes you can't use srand() and rand() from multiple threads at all. Other Unix C runtimes may behave differently.

Visual C++ runtime uses per-thread internal state, so it is safe to call srand() for each thread. But as Neil pointed out, you will likely seed all threads with same value - so seed with (time + thread-id) instead.

Of course, for portability, use Random objects rather than rand function, and then you would not depend on hidden state at all. You still need one object per thread, and seeding each object with (time + thread-id) is still a good idea.

View more solutions

Comments

[fredoverflow]

You are probably better off with the new random number generators coming in C++0x. What compiler are you using?

[ob_dev]

What OS are you using windows/linux ??

[ob_dev]

if all the treads use the same srand() you will get the same random numbers

[Johan Råde]

Don't call rand() from multiple threads. Use the random number generators in C++0x. These are also available in Boost.

[Paulo Carvalho]

I've observed that rand() starts over the same sequence everytime a thread starts. In my application I start one thread in a loop, and rand() repeated the same sequence in each iteration. rand() is definitely a no-no with multithreading. I fixed this by using those newer C++ generators like suggested above.

[Seng Cheong]

Using the memory address of a local variable is not a good source of entropy.

[AnT stands with Russia]

It is not clear how this very narrowly implementation-specific quote applies to the original question. The latter does not state what implementation is used. rand() can easily be thread safe. It is up to a specific implementation.

[Groo]

Interestingly, rand seems to be a thread-safe version of rand_r in glibc, not vice-versa (if I am interpreting the source code correctly).

[borrrden]

You are not. You are looking at exactly the internal state that makes it non thread-safe.

[omatai]

I'm going to comment "windows" just so that people more readily land on this answer when using MS Visual C++ in a Windows context.

[Nate Eldredge]

@AnT: But rand() is not required to be thread-safe, so a portable multi-threaded program cannot safely use it (except by protecting it with a mutex or similar). C17 7.22.2.1p3: "The rand function is not required to avoid data races with other calls to pseudo-random sequence generation functions [such as itself]".