By: CJ
Hi Tanel, I am confused about a trivial issue. I was reading the following excerpt from an article: A Concepual Knowledge on Oracle Blocks Read Algorithm is as under : 1. User Issues a Query. 2. Query...
View ArticleBy: Ashish
Hi Tanel, We faced the issue of library cache latch contentions due to which CPU was 0% idle continuously. There was one index rebuild session which was holding the library cache lock. Others were...
View ArticleBy: CJ
@CJ Hi Tanel, It’s been a while but this question has started bothering me again, so I thought I might ask again. Suppose a block is in the buffer cache and server process needs to fetch a row from...
View ArticleBy: Tanel Poder
@CJ CJ, The buffer is pinned (in share mode for reading) and then only the needed rows are extracted from the buffer. So no extra copy is happening. There’s an optimization (the “consistent gets –...
View ArticleBy: CJ
Brilliant, thanks a lot Tanel. If you were asked to *prove* that a copy was not happening, how would you do that? Thanks.
View ArticleBy: Tanel Poder
@CJ Use a dev instance, suspend all other processes so they couldn’t do LIOs and then run various queries on various index types (with equality predicate and range predicate) and then measure the...
View ArticleBy: CJ
@Tanel Poder Ok, I can try it, I have a local instance where the only user session is my own. I expect that I will see the increase in gets/immediate gets in v$latch to match the number of consistent...
View ArticleBy: Tanel Poder
@CJ Or regarding the copy, there’s no counter for that. You can use a debugger for that, create a block with two rows with values ADBCDEF and GHIJKLM in them. Then run a query which selects only one of...
View ArticleBy: Tanel Poder
@Tanel Poder Make sure that your test is clean – start up a completely new process for your final test etc
View ArticleBy: CJ
Or maybe I will just take your word for it ;) Seriously, I will give it a try. Thanks a lot.
View Article
