Check CPU load on multiple cores in a multi-threaded Java process

Test.java

import java.lang.management.ManagementFactory;
import com.sun.management.OperatingSystemMXBean;

public class Test {
	
	public static void main(String [] args) {
		// Probably this can vary, depending on whether you run from
		// command line or from an IDE.
		final int startingThreadCount = Thread.activeCount();
		System.out.printf("Starting thread count: %d%n", startingThreadCount);
		
		int availCPUs = Runtime.getRuntime().availableProcessors();
		System.out.printf(
				"Available CPUs: %d%n",
				availCPUs);
		int threads = availCPUs * 2;
		System.out.printf("Starting %d threads...%n", threads);
		for (int i = 0; i < threads; i++) {
			Task t = new Task();
			new Thread(t).start();
		}
		
		Thread monitor = new Thread(new Runnable() {
			public void run() {
				OperatingSystemMXBean OS = (OperatingSystemMXBean) ManagementFactory.getOperatingSystemMXBean();
				while(Thread.activeCount() > startingThreadCount + 1) {
					System.out.println("Still processing...");
					System.out.printf(
							"Active thread count: %d%n",
							Thread.activeCount());
					System.out.printf(
							"CPU load (of all CPUs): %.2f%%%n",
							OS.getProcessCpuLoad() * 100);
					try {
						Thread.sleep(1000);
					} catch (InterruptedException e) {
						// TODO Auto-generated catch block
						e.printStackTrace();
					}
				}
				System.out.println("Done!");
 			}
			
		});
		monitor.start();
	}
}

class Task implements Runnable {
	public static final long SAMPLES = 100_000_000L;
	/**
	 * Simple pi approximation based on the Leibniz formula for pi.
	 * See http://en.wikipedia.org/wiki/Leibniz_formula_for_pi
	 */
	public void run() {
		double pi = 0.0;
		
		// ten million samples
		for(long k = 0; k < SAMPLES; k++) {
			pi += 4.0 * (Math.pow(-1.0, k) / ((2.0 * k) + 1));
		}
		System.out.printf(
				"Thread: %s. Result: %f%n",
				Thread.currentThread().getName(),
				pi);
	}
}