考虑这样一个问题:统计某个工程的代码行数。首先想到的思路便是,递归文件树,每层递归里,循环遍历父文件夹下的所有子文件,如果子文件是文件夹,那么再对这个文件夹进行递归调用。于是问题很轻松的解决了。这个方案可以优化吗? 了
再回想这个问题,可以发现,循环里的递归调用其实相互之间是独立的,互不干扰,各自统计自己路径下的代码文件的行数。于是,发现了这个方案的可优化点——利用线程池进行并行处理。于是一个串行的求解方案被改进成了并行方案。
不能光说不练,写了一个Demo,对串行方案和并行方案进行了量化对比。代码如下:
import java.io.*; import java.util.Queue; import java.util.concurrent.*; /** * Created by cdlvsheng on 2016/5/16. */ public class ParallelSequentialContrast { int coreSize = Runtime.getRuntime().availableProcessors(); ThreadPoolExecutor exec = new ThreadPoolExecutor(coreSize * 4, coreSize * 5, 0, TimeUnit.SECONDS, new LinkedBlockingQueue<Runnable>(10000), new ThreadPoolExecutor.CallerRunsPolicy()); Queue<Future<Integer>> queue = new ConcurrentLinkedQueue<Future<Integer>>(); private int countLineNum(File f) { if (!f.getName().endsWith("java") && !f.getName().endsWith(".js") && !f.getName().endsWith(".vm")) return 0; int sum = 0; try { BufferedReader br = new BufferedReader(new FileReader(f)); String str = null; while ((str = br.readLine()) != null) sum++; } catch (FileNotFoundException e) { e.printStackTrace(); } catch (IOException e) { e.printStackTrace(); } return sum; } private class Task implements Callable<Integer> { File f; public Task(File f) { this.f = f; } public Integer call() throws Exception { int sum = 0; if (f.isDirectory()) { File[] fs = f.listFiles(); for (File file : fs) { if (file.isDirectory()) queue.add(exec.submit(new Task(file))); else sum += countLineNum(file); } } else sum += countLineNum(f); return sum; } } public int parallelTraverse(File f) { queue.add(exec.submit(new Task(f))); int sum = 0; while (!queue.isEmpty()) { try { Future<Integer> future = queue.poll(); sum += future.get(); } catch (InterruptedException e) { e.printStackTrace(); } catch (ExecutionException e) { e.printStackTrace(); } } exec.shutdown(); return sum; } public int sequentialTraverse(File f) { int sum = 0; if (f.isDirectory()) { File[] fs = f.listFiles(); for (File file : fs) { if (file.isDirectory()) sum += sequentialTraverse(file); else sum += countLineNum(file); } } else sum += countLineNum(f); return sum; } public void parallelTest(ParallelSequentialContrast psc, String pathname) { long start = System.currentTimeMillis(); int sum = psc.parallelTraverse(new File(pathname)); long duration = System.currentTimeMillis() - start; System.out.println(String.format("parallel test, %d lines of code were found, time cost is %d ms", sum, duration)); } public void sequentialTest(ParallelSequentialContrast psc, String pathname) { long start = System.currentTimeMillis(); int sum = psc.sequentialTraverse(new File(pathname)); long duration = System.currentTimeMillis() - start; System.out.println(String.format("sequential test, %d lines of code were found, time cost is %d ms", sum, duration)); } public static void main(String[] args) { ParallelSequentialContrast psc = new ParallelSequentialContrast(); String pathname = "D:\\Code_Git"; psc.sequentialTest(psc, pathname); psc.parallelTest(psc, pathname); } }
因为要不断的扫磁盘(虽然我的是固态硬盘),所以并行方案的线程池开的很大。IO密集型程序的相对CPU密集型程序的线程池会更大。
程序运行结果如下:
sequential test, 415079 lines of code were found, time cost is 364 ms parallel test, 415079 lines of code were found, time cost is 163 ms
可以发现,在结果同等精确的情况下,串行方案耗时是并行方案的两倍多。这个是在我个人PC上做的测试,如果是线上服务器运行,恐怕差距只会更加明显。
如果一个大任务,由许多个相互独立的子任务组成,我们就可以在这里找突破点,把一个串行程序并行化,榨干多和服务器的性能!
JDK1.7提供了一个Fork/Join的框架,其原理与这个并行方案如出一辙。Fork/Join框架在每fork一个任务后,都会把这个任务甩进一个工作队列,供线程池消费。所谓框架,也就是把常见问题的解决方案模板化,傻瓜化。关于Fork/Join框架,在我的前面一篇博客里有介绍:点击打开链接
时间: 2024-10-10 07:03:17