Re: Redo block size...

"Noons" <wizofoz2k_at_yahoo.com.au> wrote in message news:<401471f0$3$19709$afc38c87_at_news.optusnet.com.au>...
> "Yong Huang" <yong321_at_yahoo.com> wrote in message
> news:b3cb12d6.0401251443.e217f7b_at_posting.google.com...
>
> > filesystem-based. In my case, it's using Veritas Quick I/O but
> > reported as raw.
>
> That would be because the partition is managed by Veritas itself
> from a Unix raw partition?

Precisely.

> > Most OSes use 512 bytes as logfile block size. But HPUX uses 1k,
> > probably DEV_BSIZE defined in /usr/include/sys/param.h. This size is
> > actually not filesystem dependent.
>
> Actually, if the redo log file is in a file system, write size is
> totally dependent on it. If you request a write of 512 bytes in
> a file system of 4K block size, the OS *will* write 4K.
> Like it or not...
>
> And the main reason for the usual recomendation to keep redo
> log in raw.

We both got it right. Oracle's logfile block size for most OSes remains at 512 bytes as shown by dbfsize or x$kccle.lebsz or v$archived_log.block_size, regardless whether it's filesystem-based or not. But if it's on a fs, a partial block write is handled by fs as a full block read, filled with redo entries, and then a full block write. Hence the raw for logfile recommendation. For performance reason, I can't imagine anybody recommends otherwise.

Yong Huang Received on Mon Jan 26 2004 - 11:48:49 CST