|
原帖由 stacyruirui 于 2008-1-7 23:40 发表 ![]()
Is the parameter helpful? cr request event is caused by remote cache access to old image or remote instance undo block.
I'm not sure if setting _serial_direct_read to true helps reduce global cache cr requests. But in theory, it should work. It basically changes your RAC database to behave like a non-cache-fusion database, temporarily for that type of scan. That is, when a block is needed (for scattered read only; note the word "scattered" here corresponds to "serial" in "_serial_direct_read"), Oracle doesn't bother to check buffer cache; instead the process directly goes to disk and reads. If that's the case, global cr requests should drop.
On the other hand, you may ask: Whenever a direct read is needed, it's always preceded by writing dirty buffers for the needed blocks to disk first. So the *old* blocks for this segment in global buffer cache have to be scanned anyway. So we don't gain anything. Is that why you're asking?
So, I actually don't know. Somebody else may have more insight, and possibly come up with a good experiment to prove the effectiveness or lack of. I would appreciate that.
Yong Huang |
|