AWR中Avg global cache current block pin time 高的原因疑问？

fluttersnow

看了maclean 的 鹰眼第二讲后，找了以前的awr再看一遍，发现一份报告中

OS REDHAT 5.5
DB 10.2.0.4

pin time和 flush time都异常高

1. Global Cache and Enqueue Services - Workload Characteristics
   
2. 
   
3. Avg global enqueue get time (ms):         0.0
   
4. Avg global cache cr block receive time (ms):         0.4
   
5. Avg global cache current block receive time (ms):         0.4
   
6. Avg global cache cr block build time (ms):         0.0
   
7. Avg global cache cr block send time (ms):         0.0
   
8. Global cache log flushes for cr blocks served %:         1.7
   
9. Avg global cache cr block flush time (ms):         15.0
   
10. Avg global cache current block pin time (ms):         1,485.5
    
11. Avg global cache current block send time (ms):         0.0
    
12. Global cache log flushes for current blocks served %:         1.2
    
13. Avg global cache current block flush time (ms):         78.5
    

复制代码

top 5中
Log archive I/O,查询解释如下：
Log archive I/O Used local archiving of online redo logs (for a production database) or standby redo logs (for a standby database). When the archiving process exhausts its I/O buffers because all of them are being used for on-going I/O's, the wait for an available I/O buffer is captured in this system wait event.
Wait Time: Depends on the speed of the disks

1. Top 5 Timed Events
   
2. 
   
3. Event        Waits        Time(s)        Avg Wait(ms)        % Total Call Time        Wait Class
   
4. CPU time                  1,707                  84.7         
   
5. Log archive I/O         9,623         329         34         16.3        System I/O
   
6. log file parallel write         30,059         306         10         15.2        System I/O
   
7. log file sync         20,148         240         12         11.9        Commit
   
8. log file sequential read         10,020         146         15         7.2        System I/O
   

复制代码

1. 
   
2. Statistic        Total        per Hour
   
3. log switches (derived)        21        20.93
   

复制代码

1小时切了21次归档。

疑问：
是磁盘读写问题导致pin time 和 flush time有很高的值吗？

fluttersnow

整理下，原因是否是如下：
   redo量增加，而磁盘读写性能不佳，导致redo 的写入需要等待，而LMS在传输block前需要等待redo写入，所以造成cluster的几项指标很高。
   
PS： 这些指标的标准数值，如 “System I/O达到13ms ”有什么文档可以参考吗？还是一个经验值呢？

Maclean Liu(刘相兵

虽然cluster wait不是主要问题， 但IO的差 可见一斑了

Maclean Liu(刘相兵

你的这个快照 负载不高 ，
但

Log archive I/O         9,623         329         34         16.3        System I/O
log file parallel write         30,059         306         10         15.2        System I/O
log file sync         20,148         240         12         11.9        Commit
log file sequential read         10,020         146         15         7.2        System I/O

这几个物理读写的 Io指标均不佳

wait class中 System I/O达到13ms
System I/O        64,023        0.00        819        13        2.39