1.软件版本
MemSQL 3.1
Pivotal SQLFire 1.1.2
Oracle 11g XE
2.测试环境
宿主机
OS：Windows 7 64位 专业版
CPU：i7-3770 3.4GHz
内存：16G
硬盘：大于800G
VMware虚拟机
OS：CentOS 6.3 64位
内存：16G
其中MemSQL运行于虚拟机，SQLFire和Oracle直接运行于宿主机
SQLFire只开启一个server member
3.DDL
MemSQL
DROP TABLE t_main; DROP TABLE t_rel1; DROP TABLE t_rel2; DROP TABLE t_rel3; create table t_main( rec_id BIGINT not null PRIMARY KEY, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) ; create table t_rel1( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) ; create table t_rel2( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) ; create table t_rel3( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) ; create index rel1main on t_rel1(main_id); create index rel2main on t_rel2(main_id); create index rel3main on t_rel3(main_id);
SQLFire
DROP TABLE t_main; DROP TABLE t_rel1; DROP TABLE t_rel2; DROP TABLE t_rel3; create table t_main( rec_id BIGINT not null PRIMARY KEY, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) PERSISTENT; create table t_rel1( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) PERSISTENT; create table t_rel2( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) PERSISTENT; create table t_rel3( rec_id BIGINT not null PRIMARY KEY, main_id BIGINT, c1 DOUBLE PRECISION, c2 BIGINT, c3 VARCHAR(100), c4 VARCHAR(100), c5 VARCHAR(100), c6 VARCHAR(100) ) PERSISTENT; create index rel1main on t_rel1(main_id); create index rel2main on t_rel2(main_id); create index rel3main on t_rel3(main_id);
Oracle
DROP TABLE t_main; DROP TABLE t_rel1; DROP TABLE t_rel2; DROP TABLE t_rel3; create table t_main( rec_id NUMBER(19,0) not null PRIMARY KEY, c1 DOUBLE PRECISION, c2 NUMBER(19,0), c3 VARCHAR2(100), c4 VARCHAR2(100), c5 VARCHAR2(100), c6 VARCHAR2(100) ) ; create table t_rel1( rec_id NUMBER(19,0) not null PRIMARY KEY, main_id NUMBER(19,0), c1 DOUBLE PRECISION, c2 NUMBER(19,0), c3 VARCHAR2(100), c4 VARCHAR2(100), c5 VARCHAR2(100), c6 VARCHAR2(100) ) ; create table t_rel2( rec_id NUMBER(19,0) not null PRIMARY KEY, main_id NUMBER(19,0), c1 DOUBLE PRECISION, c2 NUMBER(19,0), c3 VARCHAR2(100), c4 VARCHAR2(100), c5 VARCHAR2(100), c6 VARCHAR2(100) ) ; create table t_rel3( rec_id NUMBER(19,0) not null PRIMARY KEY, main_id NUMBER(19,0), c1 DOUBLE PRECISION, c2 NUMBER(19,0), c3 VARCHAR2(100), c4 VARCHAR2(100), c5 VARCHAR2(100), c6 VARCHAR2(100) ) ; create index rel1main on t_rel1(main_id); create index rel2main on t_rel2(main_id); create index rel3main on t_rel3(main_id);
4.测试数据
4张表，每张100万条数据
生成数据的代码片段为
String sql = "insert into t_main(rec_id,c1,c2,c3,c4,c5,c6) values(?,?,?,?,?,?,?)";
PreparedStatement st1 = conn.prepareStatement(sql);
sql = "insert into t_rel1(rec_id,main_id,c1,c2,c3,c4,c5,c6) values(?,?,?,?,?,?,?,?)";
PreparedStatement st2 = conn.prepareStatement(sql);
sql = "insert into t_rel2(rec_id,main_id,c1,c2,c3,c4,c5,c6) values(?,?,?,?,?,?,?,?)";
PreparedStatement st3 = conn.prepareStatement(sql);
sql = "insert into t_rel3(rec_id,main_id,c1,c2,c3,c4,c5,c6) values(?,?,?,?,?,?,?,?)";
PreparedStatement st4 = conn.prepareStatement(sql);
long a = System.currentTimeMillis();
for (int i = 1; i <= count; i++) {
st1.setLong(1, i);
st1.setDouble(2, i * 0.1);
st1.setLong(3, i * 2);
st1.setString(4, "c3_" + i);
st1.setString(5, "c4_" + i);
st1.setString(6, "c5_" + i);
st1.setString(7, "c6_" + i);
st1.executeUpdate();
st2.setLong(1, i);
st2.setLong(2, i);
st2.setDouble(3, i * 0.2);
st2.setLong(4, i * 3);
st2.setString(5, "c3_" + i);
st2.setString(6, "c4_" + i);
st2.setString(7, "c5_" + i);
st2.setString(8, "c6_" + i);
st2.executeUpdate();
st3.setLong(1, i);
st3.setLong(2, i);
st3.setDouble(3, i * 0.3);
st3.setLong(4, i * 4);
st3.setString(5, "c3_" + i);
st3.setString(6, "c4_" + i);
st3.setString(7, "c5_" + i);
st3.setString(8, "c6_" + i);
st3.executeUpdate();
st4.setLong(1, i);
st4.setLong(2, i);
st4.setDouble(3, i * 0.4);
st4.setLong(4, i * 5);
st4.setString(5, "c3_" + i);
st4.setString(6, "c4_" + i);
st4.setString(7, "c5_" + i);
st4.setString(8, "c6_" + i);
st4.executeUpdate();
}
5.测试结果
4张表，每张插入100万数据，消耗时间对比
单位：毫秒
| MemSQL | SQLFire | Oracle |
| 624765 | 196140 | 1289811 |
以下为查询测试，均执行10次求得平均消耗时间（不包含首次执行）
单位：毫秒
查询测试一：单表整型字段比较
select count(*) from t_main where c2>1000;
| MemSQL | SQLFire | Oracle |
| 21 | 675 | 58 |
查询测试二：单表like
select count(*) from t_main where c4 like '%c%';
| MemSQL | SQLFire | Oracle |
| 41 | 875 | 133 |
查询测试三：多表关联浮点数sum
select sum(m.c1+r1.c1+r2.c1+r3.c1) "rt" from t_main m,t_rel1 r1,t_rel2 r2,t_rel3 r3 where m.rec_id=r1.main_id and m.rec_id=r2.main_id and m.rec_id=r3.main_id;
| MemSQL | SQLFire | Oracle |
| 1365 | 14640 | 2077 |
查询测试四：多表关联整型sum
select sum(m.c2+r1.c2+r2.c2+r3.c2) "rt" from t_main m,t_rel1 r1,t_rel2 r2,t_rel3 r3 where m.rec_id=r1.main_id and m.rec_id=r2.main_id and m.rec_id=r3.main_id;
| MemSQL | SQLFire | Oracle |
| 1360 | 10257 | 2084 |
6.总结
测试过程中CPU、内存使用均未超过50%
插入性能SQLFire最高，MemSQL其次，Oracle最慢，MemSQL效率约是Oracle的两倍
查询性能MemSQL最高，Oracle其次，SQLFire最慢（慢的出奇。。。），MemSQL效率约是Oracle的两倍
不知道怎样的环境能测出30倍性能的提升。。。
时间： 2024-08-12 21:55:37