ThreadGroup added to the fun with Sun Fire T1000      The program in the previous blog had limitations with the number of threads argument. Giving a large number ( not much, even something like 10000 ) was resulting in program running out of memory as it had to spawn off so many Thread objects at a time. So I thought let me add some thread pooling. The need was not to have a sophisticated and accurate pooling mechanism but a soft control limit that would prevent the program running out of memory. I felt the java class ThreadGroup was just right for that.

This is how the class looks like now:

public class TestCPU implements Runnable {
    static int  threads,computes,threadLimit;
    static boolean io;
    static boolean threading;

    public static void main (String args[]) {
        TestCPU t = new TestCPU();

        try {
            threads = (new Integer(args[0])).intValue();
            computes = (new Integer(args[1])).intValue();

            if ( args[2].equals("Y") ) io = true;
            if ( args[3].equals("Y") ) threading = true;

            threadLimit = (new Integer(args[4])).intValue();
        } catch (Exception nfe) {
            // either NumberFormat or ArrayIndexOutOfBounds
            System.err.println("Using defaults");
            threads=100;
            computes=100;
            io = false;
            threading = false;
            threadLimit = 100;
        }

        System.out.println("Threads: " + threads + " Computes: " + computes );
        System.out.println("IO: " + io );
        System.out.println("Threading: " + threading);
        System.out.println("Thread Pool Size: " + threadLimit);

        ThreadGroup pool = new ThreadGroup("pool");

        for (int i=0;i<threads;i++) {
            if ( threading ) {
                if ( pool.activeCount() < (threadLimit) ) {
                    (new Thread(pool, t)).start();
                } else {
                    --i;
                }
            } else {
                compute();
            }
        }

        System.out.println("done !");
    }

    public void run() {
        compute();
    }

    static void compute() {
        int load;
        for (int j=1; j<computes;j++)
          for (int k=1; k<computes;k++) {
            // do some computing
            load = threads + computes;

            if ( io ) System.out.println("Load: " + load);
          }
    }
}

And the time output suggests that it was over pretty quickly. Besides the elapsed time which may or may not be good enough evidence of T1's superior performance, what I observe is the %CPU usage of T1 never went beyond 3%, while the other single CPU solaris or non solaris systems stopped responding as a high CPU % was given to this java process only.

With the ThreadGroup, the program goes very close to the real life long running applications as far as the threading goes. And now I have started getting a feel of T1's advantages over the traditional CPUs. My next step would be to try running some real applications doing a huge number of concurrent transactions. But as of now, I am too impressed with this T1000 !!



Posted by chiplunkar ( Sep 21 2006, 07:00:00 AM IST ) Permalink Comments [0]