今天进行了InfluxDB和MySQL的对比测试,这里记录下结果,也方便我以后查阅。
操作系统: CentOS6.5_x64
InfluxDB版本 : v1.1.0
MySQL版本:v5.1.73
CPU : Intel(R) Core(TM) i5-2320 CPU @ 3.00GHz
内存 :12G
硬盘 :SSD
初始化SQL语句:
CREATE DATABASE testMysql; CREATE TABLE `monitorStatus` ( `system_name` VARCHAR(20) NOT NULL, `site_name` VARCHAR(50) NOT NULL, `equipment_name` VARCHAR(50) NOT NULL, `current_value` DOUBLE NOT NULL, `timestamp` BIGINT(20) NULL DEFAULT NULL, INDEX `system_name` (`system_name`), INDEX `site_name` (`site_name`), INDEX `equipment_name` (`equipment_name`), INDEX `timestamp` (`timestamp`) ) ENGINE=InnoDB;
单写测试代码(insertTest1.c):
#include <stdlib.h> #include <stdio.h> #include <time.h> #include "mysql/mysql.h" #define N 100 int main() { MYSQL *conn_ptr; int res; int t,i,j; int64_t tstamp = 1486872962; srand(time(NULL)); t=0; conn_ptr = mysql_init(NULL); if (!conn_ptr) { printf("mysql_init failed/n"); return EXIT_FAILURE; } conn_ptr = mysql_real_connect(conn_ptr,"localhost","root","","testMysql",0,NULL,0); if (conn_ptr) { for(i=1;i<= 10000;i++) { mysql_query(conn_ptr,"begin"); for(j=0;j<N;j++,t++) { char query[1024]={0}; sprintf(query,"insert into monitorStatus values ('sys_%d','s_%d','e_%d','0.%02d','%lld');", //j%10,(t+i)%10,(t+j)%10,(t+i+j)%100,tstamp); j%10,(t+i)%10,(t+j)%10,rand()%100,tstamp); //printf("query : %s/n",query); res = mysql_query(conn_ptr,query); if (!res) { //printf("Inserted %lu rows/n",(unsigned long)mysql_affected_rows(conn_ptr)); } else { fprintf(stderr, "Insert error %d: %sn",mysql_errno(conn_ptr),mysql_error(conn_ptr)); } if(j%10 == 0) tstamp+=1; } mysql_query(conn_ptr,"commit"); //printf("i=%d/n",i); } } else { printf("Connection failed/n"); } mysql_close(conn_ptr); return EXIT_SUCCESS; } View Code
可根据情况调整测试代码中的N参数。
单读测试代码(queryTest1.c):
#include <stdio.h> #include <stdlib.h> #include "mysql/mysql.h" int main() { MYSQL *conn_ptr; MYSQL_RES *res_ptr; MYSQL_ROW sqlrow; MYSQL_FIELD *fd; int res, i, j; conn_ptr = mysql_init(NULL); if (!conn_ptr) { return EXIT_FAILURE; } conn_ptr = mysql_real_connect(conn_ptr,"localhost","root","","testMysql", 0, NULL, 0); if (conn_ptr) { res = mysql_query(conn_ptr,"select * from `monitorStatus` where system_name='sys_8' and site_name='s_9' and equipment_name='e_6' order by timestamp desc limit 10000;"); if (res) { printf("SELECT error:%s/n",mysql_error(conn_ptr)); } else { res_ptr = mysql_store_result(conn_ptr); if(res_ptr) { printf("%lu Rows/n",(unsigned long)mysql_num_rows(res_ptr)); j = mysql_num_fields(res_ptr); while((sqlrow = mysql_fetch_row(res_ptr))) { continue; for(i = 0; i < j; i++) printf("%s/t", sqlrow[i]); printf("/n"); } if (mysql_errno(conn_ptr)) { fprintf(stderr,"Retrive error:s/n",mysql_error(conn_ptr)); } } mysql_free_result(res_ptr); } } else { printf("Connection failed/n"); } mysql_close(conn_ptr); return EXIT_SUCCESS; } View Code
Makefile文件:
all: gcc -g insertTest1.c -o insertTest1 -L/usr/lib64/mysql/ -lmysqlclient gcc -g queryTest1.c -o queryTest1 -L/usr/lib64/mysql/ -lmysqlclient clean: rm -rf insertTest1 rm -rf queryTest1
磁盘空间占用查询:
使用du方式(新数据库,仅为测试):
du -sh /var/lib/mysql
查询特定表:
use information_schema; select concat(round(sum(DATA_LENGTH/1024/1024), 2), 'MB') as data from TABLES where table_schema='testMysql' and table_name='monitorStatus';
测试结果:
100万条数据
[root@localhost mysqlTest]# time ./insertTest1 real 1m20.645s user 0m8.238s sys 0m5.931s [root@localhost mysqlTest]# time ./queryTest1 10000 Rows real 0m0.269s user 0m0.006s sys 0m0.002s
原始数据 : 28.6M
du方式 : 279MB
sql查询方式: 57.59MB
写入速度: 12398 / s
读取速度: 37174 / s
root@localhost mysqlTest]# time ./insertTest1 real 7m15.003s user 0m48.187s sys 0m33.885s [root@localhost mysqlTest]# time ./queryTest1 10000 Rows real 0m6.592s user 0m0.005s sys 0m0.002s
原始数据 : 286M
du方式 : 2.4G
sql查询方式: 572MB
写入速度: 22988 / s
读取速度: 1516 / s
[root@localhost mysqlTest]# time ./insertTest1 real 20m38.235s user 2m21.459s sys 1m40.329s [root@localhost mysqlTest]# time ./queryTest1 10000 Rows real 0m4.421s user 0m0.004s sys 0m0.004s
原始数据 : 858M
du方式 : 7.1G
sql查询方式: 1714MB
写入速度: 24228 / s
读取速度: 2261 / s
需要将InfluxDB的源码放入 go/src/github.com/influxdata 目录
单写测试代码(write1.go):
package main import ( "log" "time" "fmt" "math/rand" "github.com/influxdata/influxdb/client/v2" ) const ( MyDB = "testInfluxdb" username = "root" password = "" ) func queryDB(clnt client.Client, cmd string) (res []client.Result, err error) { q := client.Query{ Command: cmd, Database: MyDB, } if response, err := clnt.Query(q); err == nil { if response.Error() != nil { return res, response.Error() } res = response.Results } else { return res, err } return res, nil } func writePoints(clnt client.Client,num int) { sampleSize := 1 * 10000 rand.Seed(42) t := num bp, _ := client.NewBatchPoints(client.BatchPointsConfig{ Database: MyDB, Precision: "us", }) for i := 0; i < sampleSize; i++ { t += 1 tags := map[string]string{ "system_name": fmt.Sprintf("sys_%d",i%10), "site_name":fmt.Sprintf("s_%d", (t+i) % 10), "equipment_name":fmt.Sprintf("e_%d",t % 10), } fields := map[string]interface{}{ "value" : fmt.Sprintf("%d",rand.Int()), } pt, err := client.NewPoint("monitorStatus", tags, fields,time.Now()) if err != nil { log.Fatalln("Error: ", err) } bp.AddPoint(pt) } err := clnt.Write(bp) if err != nil { log.Fatal(err) } //fmt.Printf("%d task done/n",num) } func main() { // Make client c, err := client.NewHTTPClient(client.HTTPConfig{ Addr: "http://localhost:8086", Username: username, Password: password, }) if err != nil { log.Fatalln("Error: ", err) } _, err = queryDB(c, fmt.Sprintf("CREATE DATABASE %s", MyDB)) if err != nil { log.Fatal(err) } i := 1 for i <= 10000 { defer writePoints(c,i) //fmt.Printf("i=%d/n",i) i += 1 } //fmt.Printf("task done : i=%d /n",i) } View Code
单读测试代码(query1.go):
package main import ( "log" //"time" "fmt" //"math/rand" "github.com/influxdata/influxdb/client/v2" ) const ( MyDB = "testInfluxdb" username = "root" password = "" ) func queryDB(clnt client.Client, cmd string) (res []client.Result, err error) { q := client.Query{ Command: cmd, Database: MyDB, } if response, err := clnt.Query(q); err == nil { if response.Error() != nil { return res, response.Error() } res = response.Results } else { return res, err } return res, nil } func main() { // Make client c, err := client.NewHTTPClient(client.HTTPConfig{ Addr: "http://localhost:8086", Username: username, Password: password, }) if err != nil { log.Fatalln("Error: ", err) } q := fmt.Sprintf("select * from monitorStatus where system_name='sys_5' and site_name='s_1' and equipment_name='e_6' order by time desc limit 10000 ;") res, err2 := queryDB(c, q) if err2 != nil { log.Fatal(err) } count := len(res[0].Series[0].Values) log.Printf("Found a total of %v records/n", count) } View Code
查看整体磁盘空间占用:
du -sh /var/lib/influxdb/
查看最终磁盘空间占用:
du -sh /var/lib/influxdb/data/testInfluxdb
[root@localhost goTest2]# time ./write1 real 0m14.594s user 0m11.475s sys 0m0.251s [root@localhost goTest2]# time ./query1 2017/02/12 20:00:24 Found a total of 10000 records real 0m0.222s user 0m0.052s sys 0m0.009s
原始数据 : 28.6M
整体磁盘占用:27M
最终磁盘占用:21M
写入速度: 68521 / s
读取速度: 45045 / s
1000万条数据
[root@localhost goTest2]# time ./write1 real 2m22.520s user 1m51.704s sys 0m2.532s [root@localhost goTest2]# time ./query1 2017/02/12 20:05:16 Found a total of 10000 records real 0m0.221s user 0m0.050s sys 0m0.003s
原始数据 : 286M
整体磁盘占用:214M
最终磁盘占用:189M 写入速度: 70165 / s
读取速度: 45249 / s
[root@localhost goTest2]# time ./write1 real 7m19.121s user 5m49.738s sys 0m8.189s [root@localhost goTest2]# ls query1 query1.go write1 write1.go [root@localhost goTest2]# time ./query1 2017/02/12 20:49:40 Found a total of 10000 records real 0m0.233s user 0m0.050s sys 0m0.012s
原始数据 : 858M
整体磁盘占用:623M
最终磁盘占用:602M
写入速度: 68318 / s
读取速度: 42918 / s
整体磁盘占用情况对比:
最终磁盘占用情况对比:
写入速度对比:
读取速度对比:
相比MySQL来说,InfluxDB在磁盘占用和数据读取方面很占优势,而且随着数据规模的扩大,查询速度没有明显的下降。
针对时序数据来说,InfluxDB有明显的优势。
好,就这些了,希望对你有帮助。
本文github地址:
https://github.com/mike-zhang/mikeBlogEssays/blob/master/2017/ 20170212_InfluxDB读写性能测试.md
欢迎补充