Re: Bigger block sizes

From: Stefan Knecht <knecht.stefan_at_gmail.com>
Date: Fri, 2 Oct 2015 15:04:34 +0700
Message-ID: <CAP50yQ9WcDMWdEzgFW0tB5qFWr2F+NhccN9wdExHdb8QECNJfA_at_mail.gmail.com>

On Fri, Oct 2, 2015 at 3:04 AM, Jonathan Lewis <jonathan_at_jlcomp.demon.co.uk> wrote:

>
> Picking a different block size because it isolates a particular problem to
> a specifically (limited) cache is also a good supporting reason for picking
> a special block size for LOBs - I find that "cache read" for lobs is a good
> choice, but I don't want a large volume of LOB data to waste the rest of
> the buffer cache.
>
>
>

This. I use this concept with LOBs for a long time with great success. I've also measured fairly large read performance improvements reading the same data on 8k vs larger blocksizes when using LOBs, on various different systems. Both for basic and securefile LOBs, larger blocks resulted in a clearly measurable performance improvement. Of course, your mileage may vary, but it's certainly worth testing it out.

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

Received on Fri Oct 02 2015 - 10:04:34 CEST