看到分治任务立马就想到归并排序，用Fork/Join又重新实现了一遍，
 /**
  * Ryzen 1700 8核16线程 3.0 GHz
  */
 @Test
 public void mergeSort() {
     long[] arrs = new long[100000000];
     for (int i = 0; i < 100000000; i++) {
         arrs[i] = (long) (Math.random() * 100000000);
     }
     long startTime = System.currentTimeMillis();
     ForkJoinPool forkJoinPool = new ForkJoinPool(Runtime.getRuntime().availableProcessors());
     MergeSort mergeSort = new MergeSort(arrs);
     arrs = forkJoinPool.invoke(mergeSort);
     //传统递归
     //arrs = mergeSort(arrs);
     long endTime = System.currentTimeMillis();
     System.out.println("耗时：" + (endTime - startTime));
 }
 /**
  * fork/join
  * 耗时：13903ms
  */
 class MergeSort extends RecursiveTask<long[]> {
     long[] arrs;
     public MergeSort(long[] arrs) {
         this.arrs = arrs;
     }
     @Override
     protected long[] compute() {
         if (arrs.length < 2) return arrs;
         int mid = arrs.length / 2;
         MergeSort left = new MergeSort(Arrays.copyOfRange(arrs, 0, mid));
         left.fork();
         MergeSort right = new MergeSort(Arrays.copyOfRange(arrs, mid, arrs.length));
         return merge(right.compute(), left.join());
     }
 }
 /**
  * 传统递归
  * 耗时：30508ms
  */
 public static long[] mergeSort(long[] arrs) {
     if (arrs.length < 2) return arrs;
     int mid = arrs.length / 2;
     long[] left = Arrays.copyOfRange(arrs, 0, mid);
     long[] right = Arrays.copyOfRange(arrs, mid, arrs.length);
     return merge(mergeSort(left), mergeSort(right));
 }
 public static long[] merge(long[] left, long[] right) {
     long[] result = new long[left.length + right.length];
     for (int i = 0, m = 0, j = 0; m < result.length; m++) {
         if (i >= left.length) {
             result[m] = right[j++];
         } else if (j >= right.length) {
             result[m] = left[i++];
         } else if (left[i] > right[j]) {
             result[m] = right[j++];
         } else result[m] = left[i++];
     }
     return result;
 }

@王伟童鞋的问题，我们也有这场景：通过手机号查询商家信息。
我们是在 redis 里维护(手机号，商家号)关联关系，在 redis 里通过手机号查询商家号，就知道该去哪个库表查询商家具体信息了。

内存开销：
手机号，11 个字符，占用 11B；
商家号，4 个字符，占用 4B；
一条记录占用 15B，100 万条记录，就 15*100万B，大概是：
15*1,000,000B/1000/1000=15M