- UID
- 816
- 阅读权限
- 100
- 帖子
- 4826
- 精华贴数
- 3
- 技术排名
- 200
- 技术积分
- 9663
- 社区排名
- 3619
- 社区积分
- 197
- 注册时间
- 2001-10-9
- 精华贴数
- 3
- 技术积分
- 9663
- 社区积分
- 197
- 注册时间
- 2001-10-9
- 论坛徽章:
- 23
|
发表于 2005-11-20 12:04:14
|显示全部楼层
One problem with too many nodes / instances in RAC is indirectly related to the setting of MAX_COMMIT_PROPAGATION_DELAY. Contrary to Oracle's belief, many applications do require this parameter to be set to 0 to avoid read inconsistency in the application. (By this, I mean when you commit and immediately read the result, you sometimes see a result as if the previous commit was not done. I don't mean any data corruption inside the database.)
This problem is more common than Oracle thought and the only solution is setting that parameter to 0, in spite of Oracle's warning. Why is this related to number of instances of RAC? Because if you have only two instances, setting this parameter to 0 causes no or very little performance overhead. But with more instances, God forbid 128, setting it to 0 causes a very noticeable delay on each commit, no matter how fast cross-instance communication is.
Yong Huang |
|
楼主: juniper
IP卡
狗仔卡
发表于 2005-10-26 09:40:24
显身卡
