...
负载 / 查询线程数 | n = 1 (points / ms) | n = 3 (points / ms) | n = 5 (points / ms) | |||
---|---|---|---|---|---|---|
数据结构 | Array | SkipList | Array | SkipList | Array | SkipList |
单序列一千万个点 | 28.40 K | 40.76 K | 40.01 K | 114.63 K | 42.44 K | 162.16 K |
一千序列每序列一万个点 | 27.31 K | 42.31 K | 66.73 K | 96.48 K | 89.31 K | 112.91 K |
十万序列每序列一百个点 | 32.20 K | 8.40 K | 60.43 K | 28.92 K | 82.36 K | 35.22 K |
Array/Skiplist查询延迟比随序列中的点数变化趋势
Array/Skiplist查询吞吐比随序列中的点数变化趋势
结论
(1) 内存占用:跳表的内存占用为 数组的10倍左右
(2) 写延迟:跳表的写延迟大约为 数组的20倍
(3) 读延迟:由于内存拷贝及排序,内存中的点数越多,数组查询的性能越差,在1000万点时,数组的查询延迟约为跳表的5倍,100点时,数组的查询延迟约为跳表的1.5倍
实验对比图:
结论
(1)内存占用:跳表的内存占用为 数组的10倍左右
(2)写延迟:跳表的写延迟大约为 数组的20倍
(3)读延迟:由于内存拷贝及排序,内存中的点数越多,数组查询的性能越差,在1000万点时,数组的查询延迟约为跳表的5倍,100点时,数组的查询延迟约为跳表的1.5倍
(4)读写混合负载写入吞吐:随着每序列的点数增加,数组比跳表写入吞吐越来越高,从9倍升高到40倍
(5)读写混合负载查询吞吐:随着每序列的点数增加,数组的查询吞吐一开始比跳表高,超过5000点后低于跳表(4) 读写混合负载查询吞吐:随着每序列的点数减少,跳表的查询吞吐越来越低,最终会低于数组的查询吞吐
关键实验代码
单线程:
Code Block |
---|
package data_structure; import java.util.Random; public class Main { private static int timeseriesNum = 1; private static int size = 10000000; static Random random = new Random(); public static void main(String[] args) throws InterruptedException { multiWriteAndRead(); // for memory calculate Thread.sleep(100000000); } private static void multiWriteAndRead() throws InterruptedException { MemTable[] arrayTable = new MemTable[timeseriesNum]; MemTable[] skipListTable = new MemTable[timeseriesNum]; // skip list long curTime = System.nanoTime(); for (int i = 0; i < timeseriesNum; i++) { skipListTable[i] = new SkipListTable(size); write(skipListTable[i]); } long finishTime = System.nanoTime(); System.out.println("SkipList write time: " + (finishTime - curTime) / 1_000_000 + " ms."); curTime = System.nanoTime(); for (int i = 0; i < timeseriesNum; i++) { query(skipListTable[i]); } finishTime = System.nanoTime(); System.out.println("SkipList query time: " + (finishTime - curTime) / 1_000_000 + " ms."); // array curTime = System.nanoTime(); for (int i = 0; i < timeseriesNum; i++) { arrayTable[i] = new ArrayTable(size); write(arrayTable[i]); } finishTime = System.nanoTime(); System.out.println("Array write time: " + (finishTime - curTime) / 1_000_000 + " ms."); curTime = System.nanoTime(); for (int i = 0; i < timeseriesNum; i++) { query(arrayTable[i]); } finishTime = System.nanoTime(); System.out.println("Array query time: " + (finishTime - curTime) / 1_000_000 + " ms."); System.out.println("finished"); } private static int query(MemTable table){ return table.query(); } public static void write(MemTable table){ for (int i = 0; i < size; i++) { if(random.nextDouble() < 0.2){ table.add(random.nextInt(100000) - 50000); } else{ table.add(i); } } } } |
...