Tiny并行计算框架之使用介绍Tiny并行计算框架之使用介绍

夏令秋
2023-12-01
呵呵,昨天看到两新粉,一激动,就想着今天来写这篇文章。  其实一直在关注这个领域,但是一直没有信心来写,所以一直期望着有一个开源的来用。 
看到了彭渊大师的 淘宝分布式框架Fourinone介绍,确实有一种相见恨晚的感觉,于是就准备去研究一番,详细见本人的感想文章由 fourinone初步学习想到的,确实来说,感觉到有一种啃不动的感觉,当然也可能是本人水平不足的原因所致。但是不管怎么说,促动了本人来写一个简单的并行计算框架。 
在此引用本人的 名言:“牛人的代码就是生手也一看就懂;生手的代码就是牛人来了也看不懂。” 
好的,亲们,不管你是生手还是牛人,let's GO! 
HelloWorld之一 当然,还是从Hello说起,不过这次的hello与之前不太一样,管呢,先看看再说: 

1
2
3
4
5
6
7
8
9
10
11
12
13
public class WorkerHello extends AbstractWorker {
    public WorkerHello() throws RemoteException {
        super("hello");
    }

    public Warehouse doWork(Work work) throws RemoteException {
        String name = work.getInputWarehouse().get("name");
        System.out.println(String.format("id %s: Hello %s", getId(), name));
        Warehouse outputWarehouse = new WarehouseDefault();
        outputWarehouse.put("helloInfo", "Hello," + name);
        return outputWarehouse;
    }
}



首先,工人Hello继承自抽象工人,也就是说他首先得是个工人,然后呢是个Hello工人。 
在它的构造函数中,抛出一个RemoteException,表明,它是可以被远程调用的工人,在构造方法中调用super("hello"),表明这个工人是个干hello活的工人。 
既然是工人么,因此当然得做工作了。 
首先从工作的的仓库中取出一个叫name的字符串,然后控制台打一下,然后构建了一个输出的仓库,在里面放了一个helloInfo的字符串,然后返回输出仓库,工人的任务就算完成了。 
下面看看示例代码: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
public static void main(String[] args) throws IOException, ClassNotFoundException, InterruptedException {
        JobCenter jobCenter = new JobCenterLocal();

        for (int i = 0; i < 5; i++) {
            jobCenter.registerWorker(new WorkerHello());
        }

        Foreman helloForeman = new ForemanSelectOneWorker("hello");
        jobCenter.registerForeman(helloForeman);
        Warehouse inputWarehouse = new WarehouseDefault();
        inputWarehouse.put("name", "world");
        Work work = new WorkDefault("hello", inputWarehouse);


        Warehouse outputWarehouse = jobCenter.doWork(work);
        System.out.println(outputWarehouse.get("helloInfo"));
        jobCenter.stop();
    }



首先开个职业介绍所,然后构建一了5个Hello工人,放在注册到职业介绍所去。 
然后又注册了一个专门干hello活的包工头到职业介绍所,这个包工头有点特别,随便找一个hello工人来干hello这个活。 
然后,构建了一个工作,介个工作是个hello工作,它的来料仓库里放了个name是world的值。 
然后他就对职业介绍所说,你帮咱把这个活干干。 
活干完了,也没有发生异常,顺利的在结果仓库里找到了helloInfo这个值,并且从控制台打出。 
下面是运行结果: 

1
2
id 46fbffdeb18b45f28cda4617795c2a52: Hello world
Hello,world



从上面的例子当中,我们理解了下面几个概念: 

1
2
3
4
5
6
7
8
职业介绍所:JobCenter,主要用于注册工人,注册包工头,接受或处理任务;

包工头:领取工作并招募工人,完成工作,并返回结果

工人:就是我们常说的民工了,只知道来料加工,处于生态环境的低层,最后还没有得工资

工作:只有工作类型和来料仓库
仓库:用于放各种来料或成品



职业介绍所,一般来说不用写,框架已经提供;工作,一般来说不用写;工头,绝大多数不需要写,框架已经提供了若干类型工头,一般够用了;工人,一定需要写。 
自此,简单的hello并行计算就算完成了。 
HelloWorld之二 上面的hello工作完成之后,老板突发齐想,一个hello吼得声音太小了,偶想让所有的工人都帮偶齐声喊一起Hello,World,那该多壮观,当然老板有钱,说干就干: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
public static void main(String[] args) throws IOException, ClassNotFoundException {
        JobCenter jobCenter = new JobCenterLocal();
        for (int i = 0; i < 5; i++) {
            jobCenter.registerWorker(new WorkerHello());
        }
        Foreman helloForeman = new ForemanSelectAllWorker("hello");
        jobCenter.registerForeman(helloForeman);

        Warehouse inputWarehouse = new WarehouseDefault();
        inputWarehouse.put("name", "world");
        Work work = new WorkDefault("hello", inputWarehouse);
        jobCenter.doWork(work);
        jobCenter.stop();
}



当然,这次的包工头换了一下,这个包工头会找所有的工人来干活,结果如下: 

1
2
3
4
5
id 83274d8f8c194bb89d773c232e867cc4: Hello world
id 16fbf219d3cf4ba48eef23c260de509a: Hello world
id 9c17a119a4f341d68b589a503712b0f9: Hello world
id e7e3b2bdc9444a179ad62abdd35275e1: Hello world
id 4b12a1b70f5d43e2bff473382096dfbe: Hello world



老板一看,尼妈,这帮工人喊是喊完了,这声音就响过(用的是System.out)就没有了,也不知道有几个工人给喊过,包工头说哦,我没有干这收集数据的活,你想要呀,你想要就吱声呀,我加个结果合并给你: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
public static void main(String[] args) throws IOException, ClassNotFoundException {
    JobCenter jobCenter = new JobCenterLocal();
    for (int i = 0; i < 5; i++) {
        jobCenter.registerWorker(new WorkerHello());
    }
    Foreman helloForeman = new ForemanSelectAllWorker("hello", new HelloWorkCombiner());
    jobCenter.registerForeman(helloForeman);

    Warehouse inputWarehouse = new WarehouseDefault();
    inputWarehouse.put("name", "world");
    Work work = new WorkDefault("hello", inputWarehouse);
    Warehouse outputWarehouse = jobCenter.doWork(work);
    List<String> result = outputWarehouse.get("helloInfo");
    System.out.println(result.size());
    jobCenter.stop();
}



Hello结果收集器,用于把工人干的活合并成一个结果出来: 

1
2
3
4
5
6
7
8
9
10
11
12
public class HelloWorkCombiner implements WorkCombiner {

    public Warehouse combine(List<Warehouse> warehouseList) throws RemoteException {
        Warehouse warehouse = new WarehouseDefault();
        List<String> helloList = new ArrayList<String>();
        for (Warehouse w : warehouseList) {
            helloList.add((String) w.get("helloInfo"));
        }
        warehouse.put("helloInfo", helloList);
        return warehouse;
    }
}



老板终于称心如意了。 
分布式求和 老板消停了一下下,又想,偶想知道从1加到10000这个结果值是多少。但是一个计算机算,算得太慢了,能不能多几台机器帮我看看,让我早些知道结果?(仅用于说明原理,你可以理解为从1加到10000需要几个小时) 
首先造个工人: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
public class WorkerSum extends AbstractWorker {
    public WorkerSum() throws RemoteException {
        super("sum");
    }

    public Warehouse doWork(Work work) throws RemoteException {
        long start = (Long) work.getInputWarehouse().get("start");
        long end = (Long) work.getInputWarehouse().get("end");
        long sum = 0;
        for (long i = start; i <= end; i++) {
            sum += i;
        }
        Warehouse outputWarehouse = new WarehouseDefault();
        outputWarehouse.put("sum", sum);
        return outputWarehouse;
    }
}



工人从来料仓库获取开始和结束,然后计算合计值并放在输出仓库中的sum值域中。 
但是这活该怎么分给工人呢,工人算完的结果又怎么合并呢? 
这个时候,就需要搞个工作分解合并器给包工头用了: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
public class SumSplitterCombiner implements WorkSplitterCombiner {
    public List<Warehouse> split(Work work, List<Worker> workers) throws RemoteException {
        List<Warehouse> list = new ArrayList<Warehouse>();
        long start = (Long) work.getInputWarehouse().get("start");
        long end = (Long) work.getInputWarehouse().get("end");
        long count = end - start + 1;
        long step = count / workers.size();
        for (int i = 0; i < workers.size(); i++) {
            Warehouse subInputWarehouse = new WarehouseDefault();
            subInputWarehouse.put("start", step * i + start);
            if (i == workers.size() - 1) {
                subInputWarehouse.put("end", end);
            } else {
                subInputWarehouse.put("end", step * (i + 1));
            }
            list.add(subInputWarehouse);
        }
        return list;
    }


    public Warehouse combine(List<Warehouse> warehouseList) throws RemoteException {
        Warehouse outputWarehouse = new WarehouseDefault();
        long sum = 0;
        for (Warehouse w : warehouseList) {
            sum += (Long) w.get("sum");
        }
        outputWarehouse.put("sum", sum);
        return outputWarehouse;
    }
}



一共两方法,一个分解方法,一个合并方法,非常容易理解。 
万事具备,呵呵,开工: 
?

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
public class Test {
    public static void main(String[] args) throws IOException, ClassNotFoundException, InterruptedException {
        JobCenter jobCenter = new JobCenterLocal();
        JobCenter center = new JobCenterRemote();
        for (int i = 0; i < 5; i++) {
            center.registerWorker(new WorkerSum());
        }
        Foreman helloForeman = new ForemanSelectAllWorker("sum", new SumSplitterCombiner());
        center.registerForeman(helloForeman);
        Warehouse inputWarehouse = new WarehouseDefault();
        inputWarehouse.put("start", 1l);
        inputWarehouse.put("end", 10000l);
        Work work = new WorkDefault("sum", inputWarehouse);

        Warehouse outputWarehouse = center.doWork(work);
        System.out.println(outputWarehouse.get("sum"));
        jobCenter.stop();
        center.stop();
    }
}



注意,输入仓库是两个长整型数,因此,下面两句最后的值是1-10000,而不是11~100001 

1
2
inputWarehouse.put("start", 1l);
inputWarehouse.put("end", 10000l);



下面是运算输出结果:  

150005000



多阶段任务 当然,简单的任务都是一下就干完了的,复杂的工作就需要分成多个阶段进行了。不同的阶段需要的包工头或工人又都是不一定相同的。对于解决这种类型的任务,咱也有相当简单的解决办法。 
先造个工人: 

1
2
3
4
5
6
7
8
9
10
11
12
13
public class WorkerHello extends AbstractWorker {
    public WorkerHello() throws RemoteException {
        super("hello");
    }

    public Warehouse doWork(Work work) throws RemoteException {
        String name = work.getInputWarehouse().get("name");
        System.out.println(String.format("id %s: Hello %s", getId(), name));
        Warehouse outputWarehouse = new WarehouseDefault();
        outputWarehouse.put("name", name + "_1");
        return outputWarehouse;
    }
}



这个工人有点怪,每次都是给名字后面附加一个"_1",然后原样返回。别的没有啥子不同。 
EN,然后来做做一系列的工作: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
public class TestSerialWork {
    public static void main(String[] args) throws IOException, ClassNotFoundException, InterruptedException {
        JobCenter jobCenter = new JobCenterLocal();
        for (int i = 0; i < 5; i++) {
            jobCenter.registerWorker(new WorkerHello());
        }
        Foreman helloForeman = new ForemanSelectOneWorker("hello");
        jobCenter.registerForeman(helloForeman);
        Warehouse inputWarehouse = new WarehouseDefault();
        inputWarehouse.put("name", "world");
        Work work = new WorkDefault("hello", inputWarehouse);
        work.setNextWork(new WorkDefault("hello")).setNextWork(new WorkDefault("hello"));
        Warehouse warehouse = jobCenter.doWork(work);
        System.out.println(warehouse.get("name"));
        jobCenter.stop();
    }
}



与前面的例子唯一的不同就是 


1work.setNextWork(new WorkDefault("hello")).setNextWork(new WorkDefault("hello"));



这里通过指定下一工作,来建立了一个系列工作,这里定义的工作是三步,下面是运行结果: 


1
2
3
4
id 2a53a967e3b84289beb3dbaf12a7d8be: Hello world
id e3d471c27e264a1a87cf263605bfe9bd: Hello world_1
id 2a53a967e3b84289beb3dbaf12a7d8be: Hello world_1_1
world_1_1_1



运行结果与预期完全一致。 
通过序列工作的方式可以把复杂的工作分解成简单的工作,而且不同的工作可以由不同的包工头和工人来完成。 

圆周率计算   圆周率的计算一般来说是比较费时间的,详细fourinone作者在文章   http://my.oschina.net/fourinone/blog/113731?p=3#comments中已经在详细的描述,这里仅采用其文章中所述方法。  

1
2
3
4
5
6
7
8
public static void main(String[] args)
{
    double pi=0.0;
    for(double i=1.0;i<1000000001d;i++){
        pi += Math.pow(-1,i+1)/(2*i-1);
    }
    System.out.println(4*pi);
}



来计算,先创建个工人: 
?

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
public class PiWorker extends AbstractWorker {
    public PiWorker() throws RemoteException {
        super("pi");
    }

    @Override
    protected Warehouse doWork(Work work) throws RemoteException {
        long m = (Long) work.getInputWarehouse().get("start");
        long n = (Long) work.getInputWarehouse().get("end");
        double pi = 0.0d;
        for (double i = m; i < n; i++) {
            pi += Math.pow(-1, i + 1) / (2 * i - 1);
        }
        work.getInputWarehouse().put("pi", 4 * pi);
        return work.getInputWarehouse();
    }
}



再写个拆分合并器: 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
public class PiSplitterCombiner implements WorkSplitterCombiner {
    public List<Warehouse> split(Work work, List<Worker> workers) throws RemoteException {
        List<Warehouse> list = new ArrayList<Warehouse>();
        long start = (Long) work.getInputWarehouse().get("start");
        long end = (Long) work.getInputWarehouse().get("end");
        long count = end - start + 1;
        long step = count / workers.size();
        for (int i = 0; i < workers.size(); i++) {
            Warehouse subInputWarehouse = new WarehouseDefault();
            subInputWarehouse.put("start", step * i + start);
            if (i == workers.size() - 1) {
                subInputWarehouse.put("end", end);
            } else {
                subInputWarehouse.put("end", step * (i + 1));
            }
            list.add(subInputWarehouse);
        }
        return list;
    }
    public Warehouse combine(List<Warehouse> warehouseList) throws RemoteException {
        Warehouse outputWarehouse = new WarehouseDefault();
        double pi = 0d;
        for (Warehouse w : warehouseList) {
            pi += (Double) w.get("pi");
        }
        outputWarehouse.put("pi", pi);
        return outputWarehouse;
    }
}



接下来是测试类 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
public class Test {
    public static void main(String[] args) throws IOException, ClassNotFoundException, InterruptedException {
        JobCenter jobCenter = new JobCenterLocal();
        for (int i = 0; i < 10; i++) {
            jobCenter.registerWorker(new PiWorker());
        }
        Foreman helloForeman = new ForemanSelectAllWorker("pi", new PiSplitterCombiner());
        jobCenter.registerForeman(helloForeman);
        Warehouse inputWarehouse = new WarehouseDefault();
        inputWarehouse.put("start", 1l);
        inputWarehouse.put("end", 1000000001l);
        Work work = new WorkDefault("pi", inputWarehouse);

        Warehouse outputWarehouse = jobCenter.doWork(work);
        System.out.println("pi:"+outputWarehouse.get("pi"));
        jobCenter.stop();
    }
}



运行结果:   

1
2
3
4
并行计算运行结果:
time:10326ms pi:3.141592694075038
单线程计算运行结果
time:24857ms pi:3.1415926525880504



这个结果是在本人笔记本跑出来的,笔记本是4核机器,而不是4CPU机器,所以4个并行跑,并没有得到期望的1/4的时间,而是1/2.4左右的时间,因此可以得出两个结论: 
结论1:通过并行计算,确实可以缩短计算时间,更好的利用CPU资源。 
绪论2:4核和4C还是有显著差异的。 
小结 在上面的例子中,我们展示了分布式计算的使用,应该是老小兼宜,简单易懂。 
职业介绍所,工人,工头,可以在一台计算机上的,也可以都在一台计算机上。 
现在,你可以很牛掰的说,速度慢?哥给你搞个分布式计算不就快了?

 类似资料: