Oracle 性能调优学习笔记(九)-- 优化排序操作A

优化排序操作
  SQL内存使用:
     排序操作:sort, group-by ,rollup
     hash-join
     bitmap 操作(merge ,bitmap索引等)
   性能调优
    1.将数据放在缓存中.
    2.将一次扫描数据放在磁盘上
    3.将多次扫描的数据放在磁盘(尽量避免)
  PGA是一个私有内存区域,被服务器进程控制.PGA内存管理可以自动管理.
  oracle动态调整PGA内存的大小.大约为SGA的20%,最小为10MB.
  
  自动排序区管理
  动态调整SQL内存
    PGA内存可用
    SQL操作可用.
    系统负载
  动态排序区域管理
   不需要设置set*_area_size(不推荐使用):
   dba只需要设置pga_aggregate_target为初始化参数,取决分配oracle
   实例的内存大小,
      其中20%内存提供给其他进程使用.
          80%内存给Oracle实例.
       OLTP系统:
          pga_aggregate_target=(total_mem*80%)*20%
     DSS系统:
       pga_aggregate_target=(total_mem*80%)*50%
   WORKAREA_SIZE_POLICY=AUTO| MANUAL
 
     PGA资源管理
       监控PGA的视图
      v$pgastat:pga的统计信息
     over allocation count:统计PGA,过度申请(超过pga_aggregate_target)内存的次数.
                    total bytes processed:
     cache hit percentage:命令率百分比         
        extra bytes read/written:不能允许理想状态时候,读写磁盘的总的字节数.
      select * from v$pgastat where name ='cache hit percentage';
     
      v$process_memory:
      v$sysstat:
     
      v$sql_workarea_histogram
      v$sql_workarea_active
      v$sql_workarea
      v$tempseg_usage
      v$sql_plan
      v$sql

     
    查看pga的建议信息
    v$pga_target_advice
    v$pga_target_advice_histogram
   
    PGA 工作负载工作状态
    select low_optimal_size/1024 as low_kb,
         (high_optimal_size+1)/1024 as high_kb,
      round(100*optimal_executions/total_executions) as optiomal,
      round(100*onepass_executions/total_executions) as onepass,
      round(100*multipasses_executions/total_executions) as multipass
      from v$sql_workarea_histogram
      where total_executions!=0
      order by low_kb;
     
   2.查询工作区的执行情况分布图
       select low_optimal_size "Low" ,high_optimal_size "Hight" ,
                      optimal_executions "Opt",
                      onepass_executions "One" ,
                      multipasses_executions "Multi",
                      total_executions "total"
                     from v$sql_workarea_histogram;
     
       11g的查询总的内存分布情况
    select optimal_count,round(optimal_count*100/total,2) optimal_perc,
           onepass_count,round(onepass_count*100/total,2) onepass_perc,
     multipasses_count,round(multipasses_count*100/total,2) multipass_perc
     from
       ( select decode(sum(total_executions),0,1,sum(total_executions)) total,
                       sum(optimal_executions) optimal_count,
                       sum(onepass_executions) onepass_count,
                       sum(multipasses_executions) multipasses_count
        from v$sql_workarea_histogram
          where low_optimal_size>64*1024        
       )
       /
    
    v$sql_workarea_active:当前活动工作区
           查看排序操作的当预期大小大于实际大小,那么一次(onepass).
    预期大小小于实际大小,那么多次(multipass).
    select to_number(decode(sid,65535,null,sid)) sid,
      operation_type oper,trunc(expected_size/1024) esize,
      trunc(actual_mem_used/1024) mem,
      trunc(max_mem_used/1024) " max mem"
      from v$sql_workarea_active
      order by 1,2;