我有一个服务器端应用程序,我使用 VisualVM
进行分析,使用Streaming API。
然而,由于该代码中存在很多因素,我还制作了一个玩具示例来比较流式传输与映射。 我觉得有些事情可能会因为结果中存在很多随机性。
是测量吗?使用其他类型的打字机会有所作为吗?是不是有一些我不知道的多线程?
目前我正在向NUL
文件对象写入相当于dev/null
的窗口。我在high
优先级上运行此操作,以防操作系统可能影响它。
玩具示例代码:
import java.io.File;
import java.io.IOException;
import java.io.StringWriter;
import java.util.ArrayList;
import java.util.Map.Entry;
import java.util.Scanner;
import java.util.TreeMap;
import com.fasterxml.jackson.core.JsonEncoding;
import com.fasterxml.jackson.core.JsonFactory;
import com.fasterxml.jackson.core.JsonGenerator;
import com.fasterxml.jackson.databind.ObjectMapper;
public class TestStreamingMapping {
public final static int NUM_SIMULATED_CATALOGS = 10000;
public final static int CATALOG_SIZE = 1000; //1000 Items in CATALOG, 500 requests per second
public final static boolean WRITE_TO_FILE = false; //Write to file, or write to string
public final static boolean DEBUG_PRINT_100_CHAR = false; //Print out part of string to see all ok
public static final String mappingFile = "mapping.txt"; //If writing to file, where?
public static final String streamingFile = "streaming.txt"; //If streaming to file, where?
public static final boolean PRINT_INTERMEDIATE_RESULTS = false;
public static TreeMap<Long,Double> iterationPercentages = new TreeMap<Long,Double>();
ObjectMapper mapper= new ObjectMapper();
JsonFactory f = new JsonFactory();
JsonGenerator g;
public static long totalCountStream = 0, totalCountMap = 0;
public static void main(String args[])
{
System.out.println("Press enter when profiler is connected...");
new Scanner(System.in).nextLine();
System.out.println("Starting iterations of JSON generation.");
double percentage;
for(long i=0; i<NUM_SIMULATED_CATALOGS; i++)
{
performTest();
percentage = (totalCountStream*100.0d / totalCountMap);
iterationPercentages.put(i, percentage);
if(!PRINT_INTERMEDIATE_RESULTS && i%100 == 0)System.out.print(i+"-");
}
System.out.println("Total Streaming API: " + totalCountStream + " ns.");
System.out.println("Total Mapping API: " + totalCountMap + " ns.");
System.out.println("Total Stream(as % of map): " + totalCountStream*100.0d / totalCountMap + " %\r\n" );
System.out.println("Iteration\tStreamPercent");
for(Entry<Long, Double> entry : iterationPercentages.entrySet())
if(entry.getKey() % 20 ==0)
System.out.println(entry.getKey() + "\t\t" + Math.round(entry.getValue()) + "%" );
}
public static void performTest()
{
TestStreamingMapping test = new TestStreamingMapping();
long time1, time2;
double percentage = 0;
try {
long starttime1 = System.nanoTime();
test.streamingToFile();
totalCountStream+=time1=System.nanoTime() - starttime1;
long starttime2 = System.nanoTime();
test.objectMapping();
totalCountMap+=time2=System.nanoTime() - starttime2;
percentage = (time1*100.0d / time2);
if(PRINT_INTERMEDIATE_RESULTS)
{
System.out.println("Streaming API: " + time1 + " ns.");
System.out.println("Mapping API: " + time2 + " ns.");
System.out.println("Stream(as % of map): " + percentage + " %" );
System.out.println("----------------------------------------------\r\n");
}
} catch (IOException e) {
e.printStackTrace();
}
}
public String[] numbers;
public ArrayList<String> arrayList = new ArrayList<String>();
public TestStreamingMapping()
{
numbers=new String[62];
for(int i=0; i<60; i++) numbers[i] = String.valueOf(Math.random()*i);
for(int i=0; i<60; i++) arrayList.add(String.valueOf(Math.random()*i));
}
public void initializeGenerator(StringWriter writer) throws IOException
{
if(WRITE_TO_FILE)
g = f.createGenerator(new File(mappingFile), JsonEncoding.UTF8);
else
g = f. createGenerator(writer);
}
public void objectMapping() throws IOException
{
StringWriter writer = new StringWriter();
initializeGenerator(writer);
for(int j=0; j<CATALOG_SIZE; j++)
mapper.writeValue(g, this);
g.close();
writer.close();
if(DEBUG_PRINT_100_CHAR)
System.out.println(writer.toString().substring(0,100));
}
public void streamingToFile() throws IOException
{
StringWriter writer = new StringWriter();
initializeGenerator(writer);
for(int j=0; j<CATALOG_SIZE; j++)
{
g.writeStartObject();
g.writeFieldName("numbers_streaming");
g.writeStartArray();
for(int i=0; i<numbers.length; i++) g.writeString(numbers[i]);
g.writeEndArray();
g.writeFieldName("arrayList"); g.writeStartArray();
for(String num: arrayList) g.writeString(num);
g.writeEndArray();
g.writeEndObject();
}
g.close();
writer.close();
if(DEBUG_PRINT_100_CHAR)
System.out.println(writer.toString().substring(0,100));
}
}
以下代码模拟了一个服务,该服务将生成一个包含1000个Prouct对象的JSON目录文档。热点显然是产品的序列化(streamToFile()
vs objectMapping()
)。
答案 0 :(得分:2)
好的,有几件事。
最重要的是,您应该只创建一个JsonFactory
实例,类似于重用ObjectMapper
的方式。重复使用这些物体是杰克逊表现的关键因素之一。有关更多提示,请参阅here。
要考虑的另一件事是使用File
会增加I / O开销,这两种方法应该大致相同,并且会减少实际处理时间的差异。您可能希望将其分开以查看文件访问花费了多少时间。我意识到这可能是虚假的文件(根据操作系统如何处理的说明),但即使没有物理开销,操作系统通常会产生一些系统调用开销。
然后一个普遍的方面是,在测量JVM上的性能时,你总是需要记住预热时间:你应该总是将测试预热多秒(最少5或10秒),以及运行实际测试足够的时间(如30秒或更长时间),以获得更稳定的结果。 这是测试框架可以提供帮助的地方,因为它们实际上可以统计测量事物,并确定结果何时稳定到足够有意义。
希望这有帮助!