Re: reduce block size?

There's always room for a lot of argument about block sizes. However there is no absolute answer, and the potential benefit is machine and application dependent.
I once specified a 2K block size on a 4TB database because that happened to be
optimal for the available resources and the application.

2K is, for the most part, considered an undesirable size of Oracle block these
days because of the potential for overheads from devices drivers and filesystems, so on a 'short review' I would not expect
someone to be able to state confidently that it was a good idea, and would certainly expect them to supply some decent maths to justify the point.

OTOH I would not be surprised by a 'low-key' suggestion that 4K might be better than 8K, and I would only expect a brief, relatively 'low-tech' note explaining the suggestion.

In either case, I would want to know that the performance benefit was going to be worth the effort of change - there are some things which just have to be left 'wrong' because the cost of making them 'right' isn't worth it.

In this case I would probably want to see LOTs of Buffer Busy Waits and Write Complete Waits, and have it proved that they were not due to some other issue.

--

Jonathan Lewis
Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk

David Spaisman wrote in message <37D88E24.6599359A_at_intercall.net>...
>Hello:
>
>I ma working with an Oracle 8.0.4 application on NT 4 sp4. There are
>about 125 users and the applicaton is definitely transaction-based wit
>about 125 tables. This application serves users in the U.S. We have a
>similar IT group in Europe and the same application there. They
>currently have less users buit the application is expected to grow(I
>don't know at what rate)..
>
>My DBA associate in Europe had a consultant(I believe from Oracle)
>review the application and made three recomendations ;1) Take care of
>some fragmented tables 2) decrease the shared pool as it was causing
>memory swapping and 3) reduce the block size from 8k to 2k because of
>the transactional nature of the application.
>
>I agree with the first two recommendations but I am quite surprised
>about the third: reducing the block size.
>
>Has any one ever seen this type of recommendation ? Is it correct or
>not? Has any one ever done this?
Received on Fri Sep 10 1999 - 02:30:47 CDT