JAVA中四种SQL执行方式性能测试

在这个项目中，由于一些原因没有使用orm进行数据库操作，而是使用Spring的JdbcTemplate

但在上线后发现效率很低，通过性能监视器发现SQL Compilations的比率很高，程序中大量使用动态拼SQL语句的情况

于是进行了一些性能测试

代码如下：

Main.log("SQL四种执行方式性能比较");
// Statement
jt.execute("dbcc freeproccache");
start = System.currentTimeMillis();
sql = "select top {0} id, intStatus from tblInfo where charCityFrom=''{1}''";
for (int i = citys.size() -1; i >= 0; i--) {
Map c = citys.get(i);
String city = (String) c.get("charCityFrom");
jt.queryForList(MessageFormat.format(sql, i % 20 + 1, city));
}
map = jt.queryForMap("Select Count(*) CNT,sum(size_in_bytes) TotalSize From sys.dm_exec_cached_plans");
Main.log("Statement：CNT=" + map.get("CNT") + " TotalSize=" + map.get("TotalSize"), start);

// PreparedStatement
jt.execute("dbcc freeproccache");
start = System.currentTimeMillis();
sql = "select top {0} id, intStatus from tblInfo where charCityFrom=?";
for (int i = citys.size() -1; i >= 0; i--) {
Map c = citys.get(i);
String city = (String) c.get("charCityFrom");
jt.queryForList(MessageFormat.format(sql, i % 20 + 1), new Object[]{city});
}
map = jt.queryForMap("Select Count(*) CNT,sum(size_in_bytes) TotalSize From sys.dm_exec_cached_plans");
Main.log("PreparedStatement：CNT=" + map.get("CNT") + " TotalSize=" + map.get("TotalSize"), start);

// EXEC sp_executesql
jt.execute("dbcc freeproccache");
start = System.currentTimeMillis();
sql = "EXEC sp_executesql N''select top {0} id, intStatus from tblInfo where charCityFrom=@cityFrom'', N''@cityFrom nvarchar(10)'', ''{1}''";
for (int i = citys.size() -1; i >= 0; i--) {
Map c = citys.get(i);
String city = (String) c.get("charCityFrom");
jt.queryForList(MessageFormat.format(sql, i % 20 + 1, city));
}
map = jt.queryForMap("Select Count(*) CNT,sum(size_in_bytes) TotalSize From sys.dm_exec_cached_plans");
Main.log("Sp_executesql：CNT=" + map.get("CNT") + " TotalSize=" + map.get("TotalSize"), start);

// 存储过程
jt.execute("dbcc freeproccache");
start = System.currentTimeMillis();
sql = "exec sp_FindInfoByCitFrom {0},''{1}''";
for (int i = citys.size() -1; i >= 0; i--) {
Map c = citys.get(i);
String city = (String) c.get("charCityFrom");
jt.queryForList(MessageFormat.format(sql, i % 20 + 1, city));
}
map = jt.queryForMap("Select Count(*) CNT,sum(size_in_bytes) TotalSize From sys.dm_exec_cached_plans");
Main.log("Procedure：CNT=" + map.get("CNT") + " TotalSize=" + map.get("TotalSize"), start);

 得到结果（多次执行，结果相关不大）：

2008-11-05 10:36:52.521 [main] - SQL四种执行方式性能比较
2008-11-05 10:36:53.365 [main] - Statement：CNT=224 TotalSize=9134080 t=829ms
2008-11-05 10:36:53.815 [main] - PreparedStatement：CNT=23 TotalSize=901120 t=450ms
2008-11-05 10:36:55.44 [main] - Sp_executesql：CNT=26 TotalSize=1351680 t=1625ms
2008-11-05 10:36:55.864 [main] - Procedure：CNT=25 TotalSize=933888 t=424ms

存储过程虽然性能最高，但维护确实有点麻烦

综合一下就我们的情况使用PreparedStatement最好，不过以前使用sql server 2000时有发现这个性能不好，看来跟生产环境还是有挺大关系

Sp_executesql的表现倒是出乎意料

经验主意害死人啊，就这个问题浪费了大量的时间