1）Latch的作用： A、序列化访问：保护SGA中的共享数据结构；保护共享内存的分配。 B、序列化执行：避免同时执行某些关键代码；避免互相干扰。
2）Latch请求的两种类型： A、willing-to-wait：请求的进程经过短时间的等待后再次发出请求，直到获得latch B、immediate：如果没有获得latch，请求的进程不等待，而是继续处理其他指令。 2、检查Latch竞争 检查latch free是不是主要的wait event： Select * from v$system_event order by time_waited;
检查latch的使用情况： Select * from v$latch: 与willing-to-wait请求有关的列：gets、misses、sleeps、wait_time、cwait_time、spin_gets 与immediate请求有关的列：immediate_gets、immediate_misses
Gets: number of successful willing-to-wait requests for a latch; Misses: number of times an initial wiling-to-wait request was unsuccessful; Sleeps: number of times a process waited after an initial willing-to-wait request; Wait_time: number of milliseconds waited after willing-to-wait request; Cwait_time: a measure of the cumulative wait time including the time spent spinning and sleeping,the overhead of context switches due to OS time slicing and page faults and interrupts; Spin_gets: gets that misses first try but succeed after spinning.
Immediate_gets: number of successful immediate requests for each latch; Immediate_misss: number of unsuccessful immediate requests for each latch;
一般无需调整latch，但是下列的措施是有用的： A、对处于竞争中的latch做进一步的调查 B、如果竞争主要存在于shared pool和library cache中，可以考虑调整应用 C、如果进一步的调查显示需要调整shared pool和buffer cache，就进行调整
Select * from v$latch where name like ‘%shared pool%’ or name like ‘%library cache%’；
如果竞争是在shared pool或library cache上，表示下列集中情况： A、不能共享的sql,应检查他们是否相似，考虑以变量代替sql中的常量： Select sql_text from v$sqlarea where executions=1 order by upper(sql_text); B、共享sql被重新编译，考虑library cache的大小是否需要调整： SELECT sql_text,parse_calls,executions FROM v$sqlarea where parse_calls>5; C、library cache不够大。
转载于:https://www.cnblogs.com/zndavid/archive/2009/04/08/1431848.html
相关资源：ORACLE9i_优化设计与系统调整