Question

情景如此......

我有两个函数Function A（）和Function B（）。函数A（）有一个for循环，它在每次迭代中调用函数B（）。函数B（）有一个for循环，迭代大约1000万次。

Function A()
{
    for (i = 0; i < 10; i++)
        Function B();
}

Function B()
{
    for (i = 0; i < 10000000; i++)
        certain_operations()
}

现在，我面临的问题是，A（）for-loop的第一次迭代需要1分钟才能执行，第二次迭代需要2分钟，第三次迭代需要4分钟等等...即使迭代次数相同，A（）中的后续循环执行也会增加。我已经生成了以下日志。

Cluster :  1
 INFO 2014-09-17 16:36:20,496 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 16:37:11,762 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 16:37:11,762 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 1 - Critical Operations : 160000

Cluster :  2
 INFO 2014-09-17 16:37:11,767 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 16:38:02,365 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 16:38:02,365 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 2 - Critical Operations : 40000

Cluster :  3
 INFO 2014-09-17 16:38:02,366 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 16:41:35,675 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 16:41:35,675 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 3 - Critical Operations : 80000

Cluster :  4
 INFO 2014-09-17 16:41:35,676 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 16:50:17,875 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 16:50:17,875 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 4 - Critical Operations : 120000

Cluster :  5
 INFO 2014-09-17 16:50:17,886 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 17:02:42,819 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 17:02:42,819 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 5 - Critical Operations : 160000

Cluster :  6
 INFO 2014-09-17 17:02:42,820 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 17:07:01,916 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 17:07:01,916 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 6 - Critical Operations : 40000

Cluster :  7
 INFO 2014-09-17 17:07:01,917 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 17:16:21,122 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 17:16:21,122 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 7 - Critical Operations : 100000

Cluster :  8
 INFO 2014-09-17 17:16:21,123 [pool-8-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-17 17:22:40,371 [pool-8-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-17 17:22:40,371 [pool-8-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 8 - Critical Operations : 60000

在这里，打印1和打印2之间的时间是功能A（）的一次迭代的执行时间。

知道为什么会出现这个问题吗？关于如何解决这个问题的建议将不胜感激。非常感谢。

这是我的代码，万一你可以得到异常......

Function A()
{
    for(Data d : packList)              
            List<Data> region  = getRegionData(d, dataList, axisEN, eN, cluster++);         

            //dataList -> 10 million records
            //d -> single instance of dataList
}

public List<Data> getRegionData(Data data, List<Data> listData, int axisEN, int eN, int cluster)    //Function B()
    {   
        List<Data> region = new ArrayList<Data>();
        Data temp = new Data();
        int distance;

        Logger.sysLog(LogValues.info, this.getClass().getName(), "Print 1");
        for(Data d : listData)
        {
            temp.setOffnet(java.lang.Math.abs(data.getOffnet() - d.getOffnet()));
            temp.setOnnet(java.lang.Math.abs(data.getOnnet() - d.getOnnet()));
            temp.setInet(java.lang.Math.abs(data.getInet() - d.getInet()));

            distance = temp.getOffnet() + temp.getOnnet() + temp.getInet();

            if(distance <= eN)
                if(temp.getOffnet() <= axisEN && temp.getOnnet() <= axisEN && temp.getInet() <= axisEN)
                {                   
                    d.setId(listData.indexOf(d));
                    d.setCluster(cluster);
                    d.setDistance(distance);
                    d.setCount(d.getCount() + 1);
                    region.add(d);  
                }
        }



        Logger.sysLog(LogValues.info, this.getClass().getName(), "Print 2");
        Logger.sysLog(LogValues.info, this.getClass().getName(), "Cluster " + cluster + " : " + region.size());
        return region;
    }
}

。

编辑1：删除listData.indexOf（d）

 Cluster :  1
 INFO 2014-09-18 14:58:28,306 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 14:59:04,233 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 14:59:04,233 [pool-1-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 1  - Critical Operations : 120000
 Cluster :  2
 INFO 2014-09-18 14:59:04,236 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:01:32,566 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:01:32,566 [pool-2-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 2  - Critical Operations : 120000
 Cluster :  3
 INFO 2014-09-18 15:01:32,567 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:09:15,873 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:09:15,873 [pool-3-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 3  - Critical Operations : 200000
 Cluster :  4
 INFO 2014-09-18 15:09:15,874 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:13:30,019 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:13:30,019 [pool-4-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 4  - Critical Operations : 80000
 Cluster :  5
 INFO 2014-09-18 15:13:30,019 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:24:10,602 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:24:10,602 [pool-5-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 5  - Critical Operations : 160000
 Cluster :  6
 INFO 2014-09-18 15:24:10,603 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:29:47,132 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:29:47,132 [pool-6-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 6  - Critical Operations : 60000
 Cluster :  7
 INFO 2014-09-18 15:29:47,133 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 1
 INFO 2014-09-18 15:38:49,182 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Print 2
 INFO 2014-09-18 15:38:49,182 [pool-7-thread-1][] com.bng.dbscan.util.GenerateCluster - Cluster 6  - Critical Operations : 60000

PrintGCDetails进行三次迭代

After Iter1
Heap
 PSYoungGen      total 18432K, used 952K [0x9ee80000, 0xa0300000, 0xb3540000)
  eden space 15872K, 6% used [0x9ee80000,0x9ef6e248,0x9fe00000)
  from space 2560K, 0% used [0xa0080000,0xa0080000,0xa0300000)
  to   space 2560K, 0% used [0x9fe00000,0x9fe00000,0xa0080000)
 ParOldGen       total 41728K, used 0K [0x76140000, 0x78a00000, 0x9ee80000)
  object space 41728K, 0% used [0x76140000,0x76140000,0x78a00000)
 PSPermGen       total 16384K, used 1771K [0x72140000, 0x73140000, 0x76140000)
  object space 16384K, 10% used [0x72140000,0x722faf00,0x73140000)


After Iter2
Heap
 PSYoungGen      total 18432K, used 952K [0x9ef00000, 0xa0380000, 0xb35c0000)
  eden space 15872K, 6% used [0x9ef00000,0x9efee260,0x9fe80000)
  from space 2560K, 0% used [0xa0100000,0xa0100000,0xa0380000)
  to   space 2560K, 0% used [0x9fe80000,0x9fe80000,0xa0100000)
 ParOldGen       total 41728K, used 0K [0x761c0000, 0x78a80000, 0x9ef00000)
  object space 41728K, 0% used [0x761c0000,0x761c0000,0x78a80000)
 PSPermGen       total 16384K, used 1771K [0x721c0000, 0x731c0000, 0x761c0000)
  object space 16384K, 10% used [0x721c0000,0x7237af00,0x731c0000)


After Iter3
Heap
 PSYoungGen      total 18432K, used 952K [0x9ee80000, 0xa0300000, 0xb3540000)
  eden space 15872K, 6% used [0x9ee80000,0x9ef6e248,0x9fe00000)
  from space 2560K, 0% used [0xa0080000,0xa0080000,0xa0300000)
  to   space 2560K, 0% used [0x9fe00000,0x9fe00000,0xa0080000)
 ParOldGen       total 41728K, used 0K [0x76140000, 0x78a00000, 0x9ee80000)
  object space 41728K, 0% used [0x76140000,0x76140000,0x78a00000)
 PSPermGen       total 16384K, used 1771K [0x72140000, 0x73140000, 0x76140000)
  object space 16384K, 10% used [0x72140000,0x722faf00,0x73140000)

Answer 1

线索为listData.indexOf(d)，使算法复杂度呈二次方用简单的循环计数器替换indexOf并享受性能差异。

Answer 2

只是我的2美分：

这些操作可能会导致性能问题......

    List<Data> region = new ArrayList<Data>();
    Data temp = new Data();
    ...
    region.add(..);

我会尝试以下步骤：

如果可能（！），尝试一次分配区域列表
尝试像这样分配区域列表：new ArrayList<Data>(initSize)
Data temp并非真正必要，避免new尝试使用诸如float，int等原始数据。
尝试在没有Logger

Java - 循环性能下降

2 个答案: