Fisher–Yates shuffle 基本思想（Knuth shuffle ）：
To shuffle an array a of n elements (indices 0..n-1):
for i from n − 1 downto 1 do
j ← random integer with 0 ≤ j ≤ i
exchange a[j] and a[i]
JDK源代码如下：
/**
* Moves every element of the List to a random new position in the list.
*
* @param list
* the List to shuffle
*
* @throws UnsupportedOperationException
* when replacing an element in the List is not supported
*/
public static void shuffle(List list) {
shuffle(list, new Random());
}
/**
* Moves every element of the List to a random new position in the list
* using the specified random number generator.
*
* @param list
* the List to shuffle
* @param random
* the random number generator
*
* @throws UnsupportedOperationException
* when replacing an element in the List is not supported
*/
@SuppressWarnings("unchecked")
public static void shuffle(List list, Random random) {
if (!(list instanceof RandomAccess)) {
Object[] array = list.toArray();
for (int i = array.length - 1; i > 0; i--) {
int index = random.nextInt(i + 1);
if (index < 0) {
index = -index;
}
Object temp = array[i];
array[i] = array[index];
array[index] = temp;
}
int i = 0;
ListIterator it = (ListIterator) list
.listIterator();
while (it.hasNext()) {
it.next();
it.set(array[i++]);
}
} else {
List rawList = (List) list;
for (int i = rawList.size() - 1; i > 0; i--) {
int index = random.nextInt(i + 1);
if (index < 0) {
index = -index;
}
rawList.set(index, rawList.set(i, rawList.get(index)));
}
}
}
测试代码，为了确保每种情况的初始化一样，使用了多个容器：
public class javaShuffle {
public static int temp = 0;
public static long start;
public static long end;
public static void main(final String args[]) {
Object changeTemp;
List numList = new ArrayList();
List firstList = new ArrayList();
List secondList = new ArrayList();
List thirdList = new ArrayList();
List fourthList = new ArrayList();
for (int i = 1; i 0; i--) {
j = rand.nextInt(i + 1);
thirdList.set(i, thirdList.set(j, thirdList.get(i)));
}
getEndTime("third shuffle run time ");
// fourth shuffle, simulate java shuffle
getStartTime();
Random random = new Random();
if (!(fourthList instanceof RandomAccess)) {
Object[] array = fourthList.toArray();
for (int i = array.length - 1; i > 0; i--) {
int index = random.nextInt(i + 1);
if (index < 0) {
index = -index;
}
Object temp = array[i];
array[i] = array[index];
array[index] = temp;
}
int i = 0;
ListIterator it = (ListIterator) fourthList.listIterator();
while (it.hasNext()) {
it.next();
it.set((Integer) array[i++]);
}
} else {
List rawList = (List) fourthList;
for (int i = rawList.size() - 1; i > 0; i--) {
int index = random.nextInt(i + 1);
if (index < 0) {
index = -index;
}
rawList.set(index, rawList.set(i, rawList.get(index)));
}
}
getEndTime("fourth shuffle run time");
// java shuffle
getStartTime();
Collections.shuffle(numList);
getEndTime("java shuffle run time ");
}
public static void swap(int a, int b) {
javaShuffle.temp = a;
a = b;
b = javaShuffle.temp;
}
public static int getRandom(final int low, final int high) {
return (int) (Math.random() * (high - low) + low);
}
public static void getStartTime() {
javaShuffle.start = System.nanoTime();
}
public static void getEndTime(final String s) {
javaShuffle.end = System.nanoTime();
System.out.println(s + ": " + (javaShuffle.end - javaShuffle.start) + "ns");
}
}
如果数值较小，例如100000级别，则输出大概是：
first shuffle run time : 85029499ns
second shuffle run time: 80909474ns
third shuffle run time : 71543926ns
fourth shuffle run time: 76520595ns
java shuffle run time : 61027643ns
first shuffle run time : 82326239ns
second shuffle run time: 78575611ns
third shuffle run time : 95009632ns
fourth shuffle run time: 105946897ns
java shuffle run time : 90849302ns
first shuffle run time : 84539840ns
second shuffle run time: 85965575ns
third shuffle run time : 101814998ns
fourth shuffle run time: 113309672ns
java shuffle run time : 35089693ns
first shuffle run time : 87679863ns
second shuffle run time: 79991814ns
third shuffle run time : 73720515ns
fourth shuffle run time: 78353061ns
java shuffle run time : 64146465ns
first shuffle run time : 84314386ns
second shuffle run time: 80074803ns
third shuffle run time : 74001283ns
fourth shuffle run time: 79931321ns
java shuffle run time : 86427540ns
first shuffle run time : 84315523ns
second shuffle run time: 81468386ns
third shuffle run time : 75052284ns
fourth shuffle run time: 79461407ns
java shuffle run time : 66607729ns
多次运行结果可能都不一样，但是基本java自带 shuffle速度最快，第三种方式次之。而第一种方法耗时最久。
如果是10000000级别，大概如下：
first shuffle run time : 2115703288ns
second shuffle run time: 3114045871ns
third shuffle run time : 4664426798ns
fourth shuffle run time: 2962686695ns
java shuffle run time : 3246883026ns first shuffle run time : 2165398466ns
second shuffle run time: 3129558913ns
third shuffle run time : 4147859664ns
fourth shuffle run time: 2911849942ns
java shuffle run time : 4311703487ns first shuffle run time : 2227462247ns
second shuffle run time: 3279548770ns
third shuffle run time : 4704344954ns
fourth shuffle run time: 2942635980ns
java shuffle run time : 3933172427ns first shuffle run time : 2200158789ns
second shuffle run time: 3172666791ns
third shuffle run time : 4715631517ns
fourth shuffle run time: 2950817535ns
java shuffle run time : 3387417676ns first shuffle run time : 2201124449ns
second shuffle run time: 3203823874ns
third shuffle run time : 4179926278ns
fourth shuffle run time: 2913690411ns
java shuffle run time : 3571313813ns first shuffle run time : 2163053190ns
second shuffle run time: 3073889926ns
third shuffle run time : 4493831518ns
fourth shuffle run time: 2852713887ns
java shuffle run time : 3773602415ns
可以看出，第一种方法速度最快，而第四种最慢。java自带 shuffle速度也不理想。
在进行大数据处理的时候，如果使用java库效率较低时，可以考虑使用其他方式。