Re: Bigger block sizes

Those who have cited improvements only tested narrowly, for specific use-cases, as I recall. Narrow and specific operations like CTAS operations, and that was it. You shouldn't base a wide-reaching decision for an entire application platform on something as specific as a CTAS operation. So they probably didn't "gain" anything from the bigger blocksize, they merely had one positive test scenario, and never tested for other related scenarios.

I don't recall a single mention of a regression test with the bigger block size.

More to the point, I don't recall anyone specifically testing a workload that generated inter-connect cache-fusion traffic on RAC with the bigger block size.

I don't want to be that guy who screams caution at every idea. I just know that a solution should be on the same scale as the problem it is addressing.

If you have a specific problem that needs improvement, then your solution should be just as specific. Tune the SQL. Use parallel execution. Fiddle with session-level parameters.

If the proposed solution is as global and wide-reaching as database block size or tablespace block size or instance-level initialization parameters, then alarm bells should always start screaming.

On 10/6/15 12:40, Orlando L wrote:
>
> It looks like some have gained by using a bigger blocksize; some have
> run into bugs using bigger blocks, some advocate against it. That is
> the conclusion.
>
>
> On Thu, Oct 1, 2015 at 10:48 AM, Michael Cunningham
> <napacunningham_at_gmail.com <mailto:napacunningham_at_gmail.com>> wrote:
>
> Orlando,
>
> a few years ago I did some performance testing of this on a data
> warehouse. Creating the database with 32k block sizes (so I only
> had one block size) did give me better performance. I don't have
> the details with me since I letf that job, but it worked well for us.
>
> Michael
>
> On Wed, Sep 30, 2015 at 2:29 PM, Orlando L <oralrnr_at_gmail.com
> <mailto:oralrnr_at_gmail.com>> wrote:
>
> List,
> Does anyone in the list use non default blocksize of greater
> than 8K for your oracle DBs; if so, is it for warehousing/OLAP
> type applications? What advantages do you get with them; any
> disadvantage.
> Orlando.
>
>
>
>
> --
> Michael Cunningham
>
>

--
http://www.freelists.org/webpage/oracle-l