Re: Block Size Question

I've been saying it for years now: 2K is so small for most modern systems that it is bordering on the joke.

The notion some people have that 2K should be used for OLTP systems is mostly based on hearsay, old data or just plain punting.

Most UNIX systems nowadays have an internal block size (the unit in which they will ALWAYS do the I/O) of 4K. Many are now starting to go for 8K.

NT now has 4K and 8K recommended by Microsoft themselves as the cluster size for modern large disks for optimal performance IN ALL CASES. And we still hear this 2K rubbish. It might have been true 10 or more years ago, but it certainly isn't now. And the fact that other databases can only do 2K is NOT a definition of how it should be done.

My experience is just like yours: for the last 10 years, I have not seen ONE SINGLE example be it OLTP or other, of ANY ORACLE database that performed better at 2K block size. I have heard people invoking that their benchmarks prove that 2K is better. In every single case it was either old data or a benchmark that didn't do what it was supposed to do and/or measure what it was supposed to measure.

So, bottom line: on just about any modern system (with the possible exception of VMS, I don't know what their OS does nowadays), use at least 4K. I'd go for 8K, but I'm not yet sure of this (not enough data).

--
Cheers
Nuno Souto
nsouto_at_nsw.bigpond.net.au.nospam
Is there a nospam domain?
http://www.users.bigpond.net.au/the_Den Kenneth C Stahl <BluesSax_at_Unforgettable.com> wrote in message news:3785E357.50C4263A_at_Unforgettable.com...
>snip
> When Oracle 7.3 came out it became apparent that using the default block size
> (2k) should no longer be used. In the organization where I working at the time
> we built two databases on the same machine. One had 2K blocks and the other
> had 4K blocks. Both databases were loaded with the same data and indexes from
> an export file. Our tests showed that the performance of the 4K database was
> anywhere from 25% to +50% of the 2K database. Of all the tests we performed we
> did not find a single test where the 2K database out-performed the 4K
> database.
Received on Fri Jul 09 1999 - 11:18:19 CDT