Re: RAC cache fusion details

On 17 Dec 2007 10:08:31 GMT, vitalis..._at_gmail.com (Jerome Vitalis) wrote:

>Thanks to all who took the time to answer.
>
>Did anyone of you actually measure response times in different scenarios
>for the same set of SQL commands, e.g.:
>- Only one instance running
>- 2 instances and all blocks involved are shared among them
>- 3+ instances
>
>.. with different cases to test all kind of possible mechanisms caused by
>the block sharing (PI, log flush, etc.)?
>
>
>-----
>échangez opinions et commentaires dans les forums de discussion.
>http://www.usenetgratuit.com/

Hi Jerome,

I didn't but I managed to identify the impact of Cache Fusion on a Full Table Scan, or should I say the impact of a Full Table Scan on Cache Fusion.

This pertains to 9iR2, but I doubt 10g will treat this differently.

Suppose you read 8 blocks in a scattered read. This will result in 8 x 1 global cache cr request to confirm the other instance did or didn't already cache the block.

In our case, as 50 percent of the data is retrieved using Full Table Scan, this is killing performance.

Regards,

-- 

Sybrand Bakker
Senior Oracle DBA