为了分析在插入海量数据到Cassandra集群或者Oracle时的表现,也就是插入速率,我们用java程序对插入数据的用时进行了采样,最终用JFreeChart把采样结果绘制出来了。
为了公平起见,我们做了以下处理:
1.所有的循环变量都放在了循环外面
2.对于Cassandra的replication-factor设置为1,这样插入数据不需要插入额外的备份。
3.对于Oracle我们用预编译语句,这样插入操作的执行计划可以重用。
4.所有的测试都在周末进行,这样不可能有其他人去干扰这些服务器。
5.这些机器上运行的其他进程都被我kill掉了,这样保证CPU,内存的专用性。
6.在用java代码插入Cassandra记录时候,我采用了thrift API, 因为它的效率比Hector API高。
以下是实验(分两部分,一是采样部分,二是数据分析部分)
Part 1:采样:
Cassandra的采样:
我们这里依然用循环插入50W条记录,不同的是,在循环的开始和循环每10000条记录时,我们把时间戳记录在List中,最终把这个List写入文本文件(cassandra_input_sample_data.txt):
- package com.charles.cassandra.demo;
-
- import java.io.File;
- import java.io.FileWriter;
-
- import java.util.ArrayList;
- import java.util.List;
-
- import org.apache.cassandra.thrift.Cassandra;
- import org.apache.cassandra.thrift.Column;
-
- import org.apache.cassandra.thrift.ColumnParent;
- import org.apache.cassandra.thrift.ConsistencyLevel;
-
- import org.apache.cassandra.thrift.TBinaryProtocol;
-
- import org.apache.thrift.protocol.TProtocol;
- import org.apache.thrift.transport.TFramedTransport;
- import org.apache.thrift.transport.TSocket;
- import org.apache.thrift.transport.TTransport;
-
- import com.charles.cassandra.util.CassandraOperationUtil;
-
- public class CassandraClusterStressTest
- {
- public static void main(String[] args)
- throws Exception
- {
-
- TTransport tr = new TFramedTransport(new TSocket("192.168.129.34",9160));
- TProtocol proto = new TBinaryProtocol(tr);
- Cassandra.Client client = new Cassandra.Client(proto);
- tr.open();
-
- if(!tr.isOpen())
- {
- System.out.println("无法连接到服务器!");
- return;
- }
-
- System.out.println("开始压力测试,我们插入50W条数据到2节点集群中");
- System.out.println("...");
-
-
- long startTime = System.currentTimeMillis();
-
- client.set_keyspace("Charles_Stress_Test2");
- ColumnParent parent = new ColumnParent("student");
-
-
-
-
-
- String key_user_id = "a";
- String k;
- long timestamp;
- Column idColumn =null;
- Column nameColumn=null;
-
-
- List<Integer> sampleData = new ArrayList<Integer>(51);
- for(int i = 0;i < 500000;i++)
- {
- k = key_user_id + i;
- timestamp = System.currentTimeMillis();
-
-
- idColumn = new Column(CassandraOperationUtil.stringToByteBuffer("id"));
- idColumn.setValue(CassandraOperationUtil.stringToByteBuffer(i + ""));
- idColumn.setTimestamp(timestamp);
- client.insert(
- CassandraOperationUtil.stringToByteBuffer(k),
- parent,
- idColumn,
- ConsistencyLevel.ONE);
-
-
- nameColumn = new Column(CassandraOperationUtil.stringToByteBuffer("name"));
- nameColumn.setValue(CassandraOperationUtil.stringToByteBuffer("student" + i));
- nameColumn.setTimestamp(timestamp);
- client.insert(
- CassandraOperationUtil.stringToByteBuffer(k),
- parent,
- nameColumn,
- ConsistencyLevel.ONE);
-
-
- if( (i==0) || ( (i+1)%10000==0)){
- sampleData.add((int)(timestamp));
- }
- }
-
-
- long endTime = System.currentTimeMillis();
-
- long elapsedTime = endTime-startTime;
-
- System.out.println("压力测试完毕,用时: "+elapsedTime+" 毫秒");
-
-
- tr.close();
-
-
- FileWriter fw = new FileWriter(new File("cassandra_insert_sample_data.txt"));
- for(int j=0;j<sampleData.size();j++){
- fw.write(sampleData.get(j)+"\n");
- }
- fw.flush();
- fw.close();
- }
-
-
- }
最终50W条记录插入完毕:控制台显示:

而且我们打开文本文件确定这些时间戳的样本都被记录了:

当然了,因为Cassandra的存储是基于内存的,所以我们定义了一个工具类用于转换字符串和字节数组:
-
-
- package com.charles.cassandra.util;
-
- import java.io.UnsupportedEncodingException;
- import java.nio.ByteBuffer;
-
-
-
-
-
-
-
-
-
- public class CassandraOperationUtil {
-
-
-
-
-
- public static ByteBuffer stringToByteBuffer(String s) throws UnsupportedEncodingException{
-
- return ByteBuffer.wrap(s.getBytes("UTF-8"));
- }
-
-
-
-
-
- public static String byteBufferToString (ByteBuffer b) throws UnsupportedEncodingException{
-
-
- byte[] bytes = new byte[b.remaining()];
-
- b.get(bytes);
-
- return new String(bytes,"UTF-8");
- }
-
- }
Oracle的采样:
我们这里依然用循环插入50W条记录,不同的是,在循环的开始和循环每10000条记录时,我们把时间戳记录在List中,最终把这个List写入文本文件(oracle_input_sample_data.txt):
-
-
- package com.charles.cassandra.demo;
-
-
- import java.io.File;
- import java.io.FileWriter;
- import java.sql.Connection;
- import java.sql.Date;
- import java.sql.DriverManager;
- import java.sql.PreparedStatement;
- import java.sql.ResultSet;
- import java.sql.Statement;
- import java.sql.ResultSetMetaData;
- import java.sql.Timestamp;
- import java.util.ArrayList;
- import java.util.List;
-
-
-
-
-
-
-
-
-
-
-
- public class OracleStressTest {
-
-
-
-
-
-
-
-
-
-
- public static void main(String[] args){
-
- String url="jdbc:oracle:thin:@192.168.129.14:15210:ora11g";
- String username="Charles_Stress_Test1";
- String password="Charles_Stress_Test1";
-
- String sDBDriver = "oracle.jdbc.driver.OracleDriver";
-
- try{
-
- System.out.println("开始压力测试,我们以预编译的方式插入50W条数据到Oracle中");
- System.out.println("...");
-
- long startTime=System.currentTimeMillis();
-
-
- Class.forName(sDBDriver).newInstance();
- Connection conn = DriverManager.getConnection(url,username,password);
-
-
-
- String rowkey=null;
- String id=null;
- String name=null;
- Date date=null;
- String statementString="insert into Student (rowkey,id,name,create_date )values(?,?,?,?)";;
-
- PreparedStatement pstmt = conn.prepareStatement(statementString);
-
-
- List<Integer> sampleData = new ArrayList<Integer>(51);
-
- for(int i=0;i<500000;i++){
-
- long timestamp = System.currentTimeMillis();
- rowkey="a"+i;
- id=""+i;
- name="student"+i;
- date= new Date(timestamp);
-
- pstmt.setString(1,rowkey);
- pstmt.setString(2, id);
- pstmt.setString(3,name);
- pstmt.setDate(4, date);
- pstmt.execute();
-
-
- if( (i==0) || ( (i+1)%10000==0)){
- sampleData.add((int)(timestamp));
- }
- }
-
-
- pstmt.close();
- conn.close();
-
- long endTime=System.currentTimeMillis();
- long elapsedTime=endTime-startTime;
-
- System.out.println("压力测试完毕,用时: "+elapsedTime+" 毫秒");
-
-
- FileWriter fw = new FileWriter(new File("oracle_insert_sample_data.txt"));
- for(int j=0;j<sampleData.size();j++){
- fw.write(sampleData.get(j)+"\n");
- }
- fw.flush();
- fw.close();
-
- }catch(Exception e){
- System.out.println("数据库连接失败");
- e.printStackTrace();
- }
-
-
-
- }
-
-
- }
最终50W条记录插入完毕:控制台显示:

而且我们打开文本文件确定这些时间戳的样本都被记录了:

Part 2: 分析采样数据并且绘制比较图:
我们用JFreechart强大的图表制作能力来绘制比较图:
首先我们依然定义一个工具类 ParseDataUtil,它可以完成两件事情,一是从样本文件中读取数据,然后时间戳相减,最终把所有每1W条数据的耗时时间存入List<Integer>对象,二是它可以吧List<Integer>对象传递给JFreechart的数据模型:
-
-
- package com.charles.parsedata.util;
-
- import java.io.BufferedReader;
- import java.io.File;
- import java.io.FileInputStream;
- import java.io.FileNotFoundException;
- import java.io.FileReader;
- import java.io.IOException;
- import java.io.InputStreamReader;
- import java.util.ArrayList;
- import java.util.List;
-
- import org.jfree.data.category.DefaultCategoryDataset;
-
-
-
-
-
-
-
-
-
- public class ParseDataUtil {
-
-
-
-
-
-
-
-
-
-
-
- public static void addDataToDataset(DefaultCategoryDataset ds, List<Integer> datas, String seriesName) {
-
- if (datas.size() <= 0)
- return;
-
-
- Integer value = 0;
- String type = null;
-
-
- for (int i = 1; i <= datas.size(); i++) {
-
- type = i + "";
- value = datas.get(i - 1);
- ds.addValue(value, seriesName, type);
- }
-
- }
-
-
-
-
-
-
-
-
- public static List<Integer> buildSampleDataListFromFile(String fileName, int numOfRecords) {
-
-
- if (numOfRecords <= 0)
- return null;
-
- try {
-
-
- List<Integer> rawSampleData = new ArrayList<Integer>(numOfRecords);
-
-
- FileInputStream fis = new FileInputStream(fileName);
- InputStreamReader isr = new InputStreamReader(fis);
- BufferedReader br = new BufferedReader(isr);
- if (br == null) {
- System.out.println("样本文件不存在!");
- return null;
- }
-
- for (int i = 0; i < numOfRecords; i++) {
- String rawRecord = br.readLine();
- rawSampleData.add(Integer.parseInt(rawRecord));
- }
-
-
- br.close();
- isr.close();
- fis.close();
-
-
-
- int sampleDataSize = rawSampleData.size() - 1;
- List<Integer> sampleData = new ArrayList<Integer>(sampleDataSize);
-
- Integer baseTimeStamp = rawSampleData.get(0);
-
-
-
-
- for (int j = 0; j < sampleDataSize; j++) {
- int time = rawSampleData.get(j + 1) - baseTimeStamp;
- System.out.println(time);
- sampleData.add(time);
- }
-
- return sampleData;
-
- } catch (Exception ex) {
- ex.printStackTrace();
- return null;
- }
-
- }
-
- }
然后我们有个最终执行画图的类,这个类吧从原始数据分析后的数据显示在图表上,并且作为对比,吧Oracle和Cassandra集群的数据显示在同一张表上:
-
-
- package com.charles.parsedata;
-
- import java.util.ArrayList;
- import java.util.List;
-
- import javax.swing.JPanel;
-
- import org.jfree.chart.ChartFactory;
- import org.jfree.chart.ChartPanel;
- import org.jfree.chart.JFreeChart;
- import org.jfree.chart.axis.NumberAxis;
- import org.jfree.chart.plot.CategoryPlot;
- import org.jfree.chart.plot.PlotOrientation;
- import org.jfree.data.category.DefaultCategoryDataset;
- import org.jfree.ui.ApplicationFrame;
- import org.jfree.ui.RefineryUtilities;
-
- import com.charles.parsedata.util.ParseDataUtil;
-
-
-
-
-
-
-
-
-
-
- public class InsertDataStressTestDataParser extends ApplicationFrame{
-
-
-
-
-
- public InsertDataStressTestDataParser(String s) {
- super(s);
- setContentPane(createDemoLine());
- }
-
- public static void main(String[] args) {
- InsertDataStressTestDataParser fjc = new InsertDataStressTestDataParser("Cassandra&Oracle插入数据对比图");
- fjc.pack();
- RefineryUtilities.centerFrameOnScreen(fjc);
- fjc.setVisible(true);
-
- }
-
-
- public static JPanel createDemoLine(){
- JFreeChart jfreechart = createChart(createDataset());
- return new ChartPanel(jfreechart);
- }
-
-
- public static JFreeChart createChart(DefaultCategoryDataset linedataset) {
-
- JFreeChart chart = ChartFactory.createLineChart("Cassandra和Oracle插入数据对比图",
- "记录数(单位:万条)",
- "用时(毫秒)",
- linedataset,
- PlotOrientation.VERTICAL,
- true,
- true,
- false
- );
- CategoryPlot plot = chart.getCategoryPlot();
-
- NumberAxis rangeAxis = (NumberAxis) plot.getRangeAxis();
- rangeAxis.setStandardTickUnits(NumberAxis.createIntegerTickUnits());
- rangeAxis.setAutoRangeIncludesZero(true);
- rangeAxis.setUpperMargin(0.20);
- rangeAxis.setLabelAngle(Math.PI / 2.0);
-
- return chart;
- }
-
-
- public static DefaultCategoryDataset createDataset() {
- DefaultCategoryDataset ds = new DefaultCategoryDataset();
-
-
- List<Integer> data1 = ParseDataUtil.buildSampleDataListFromFile("cassandra_insert_sample_data.txt",51);
- List<Integer> data2 = ParseDataUtil.buildSampleDataListFromFile("oracle_insert_sample_data.txt",51);
- ParseDataUtil.addDataToDataset(ds, data1, "Cassandra插入数据所用时间分布图");
- ParseDataUtil.addDataToDataset(ds, data2, "Oracle插入数据所用时间分布图");
-
-
- return ds;
- }
-
-
- }
最终对比图如下:

结论:
所以我们这里很清楚的看到:
(1) 无论是Cassandra集群还是Oracle,其插入操作用时都是线性的,也就是它的平均插入速率基本是恒速。
(2) 在低配置服务器上,Cassandra集群的插入数据操作耗时要高于Oracle关系数据库。
本文转自 charles_wang888 51CTO博客,原文链接:http://blog.51cto.com/supercharles888/870876,如需转载请自行联系原作者