ex05_pool.md

Using the thread pool

Our previous examples used a single worker thread, and thus only one processor core. If we want to take full advantage of multi-core processors, we need the ability to delegate expensive computations to a pool of theads. This example demonstrates the pool thread that comes bundled with threads_a_gogo.

First, we create a pool

var Threads = require('threads_a_gogo');
var pool = Threads.createPool(3);

Then we load our fibonacci function in all the pool's threads:

function fibo(n) {
	return n > 1 ? fibo(n - 1) + fibo(n - 2) : 1;
}

pool.all.eval(fibo);

Now, we can get fibonacci numbers from our pool

We request them in reverse order, to show that longer computations (fibo(40)) run in parallel with shorter ones (fibo(39), fibo(38), ...). The results won't come out in strictly decreasing order.

var remain = 11;
for (var i = 40; i >= 30; i--) {
	// extra closure to get proper scoping on 'i'
	(function(i) {
		// dispatch each request to the first available thread
		pool.any.eval('fibo(' + i + ')', function(err, val) {
			console.log('fibo(' + i + ')=' + i);
			// destroy the pool when all results have been produced
			if (--remain == 0) console.log('bye!'), pool.destroy();
		});
	})(i);
}

Typical (*) Output

(*) Execution is non-deterministic. So order may vary.

fibo(38)=38
fibo(39)=39
fibo(37)=37
fibo(35)=35
fibo(36)=36
fibo(33)=33
fibo(34)=34
fibo(31)=31
fibo(32)=32
fibo(30)=30
fibo(40)=40
bye!