Re: DB_FILE_MULTIBLOCK_READ_COUNT

In article <331B1333.1F0F_at_pmcgettigan.demon.co.uk>,   stevep_at_pmcgettigan.demon.co.uk wrote:
>
> Kelly Young wrote:
>
> > I'm also somewhat confused by the "block" terminology used in the
> > Oracle manuals. We are running DEC Unix. It is my understanding
> > that the physical disk block is 512 bytes and the Unix (OSF/1)
> > operating system does I/O in blocks of 8192 bytes. Therefore,
> > when Oracle refers to "Operating System Blocks" I assume they are
> > referring to 8192 (not 512), correct? I've never been able to
> > confirm this.
> No. Generally when Oracle is referring to O/S blocks it is for *sizing*
> purposes, so it normally means the 512 byte disk block size, not the max
> I/O request size of your OS.

I'm not sure I agree with this. "Operating System Blocks" in my mind refer to the block size of the filesystems, which is probably 8K (this is the default for UNIX filesystems). The disk block size is not O/S specific.

>
> If you were setting, say, your Oracle block size, you would always make
> sure it was a multiple of the disk block size - 512 bytes. This is why
> you see recommended Oracle block sizes of 2K, 4K, 8K and so on.
>

true.

> The I/O size may be a limiting factor to consider when performing
> multi-block reads, as any requests which are larger than the I/O size
> for your system will just be broken up into multiple smaller requests.
> Example: 2K block (Oracle), multiblock read count 8: 8x2k = 16K read.
> The OS, with an 8K I/O limit will actually break this request down into
> 2 I/O requests, as it can't do a single 16K read, so you will get no
> advantage by setting large multi-block read counts.

This is also true. In most UNIX systems, 64K is the max io size, in spite of the 8K filesystem block size.

>
> Always check your OS specific Oracle installation guide for more info.
>
> Steve Phelan.

Brenda Muller

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet