Re: DBWR write chunk

Well before 8.1 DBWR could be triggered to write in several situations, it was either the foregrounds o DBWR on timeouts that could decide.

Anjo.

Jonathan Lewis wrote:

> Things have changed since 8.0.5, I see now that
> the thing I was referring to was
> kcbswc - DBWR max outstanding writes (8.1 version)
> formerly known as
> kcbswc - DB writer IO clump. (8.0 version)
>
> But Oracle 8.1 has introduced
> kcbscw - DBWR write chunk.
>
> However, kcbswc as was had two functions - one to
> limit the number of blocks that were allowed to be
> dirty and the other to specify the number of blocks that
> DBWR would attempt to write.
>
> i.e. 8.0 used to work on the basis that DBWR had
> to write if it found kcbswc blocks dirty, and the number
> of blocks it had to try to write if cued was also kcbswc,
> and this used to default (sometimes through bugs) to
> a very large value and cause DBWR to get too busy.
>
> I think that these two functions have now been separated:
> DBWR has to write if it finds that kcbswc blocks are dirty,
> but the number of blocks that may be written at any one
> time is limited to the smaller kcbscw value.
>
> As you indicate, _db_block_write_batch has disappeared
> in 8.1 to be split into two parameters corresponding to
> the two x$kvii tags.
>
> Thanks for the update.
>
> --
> Jonathan Lewis
> Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk
>
> Practical Oracle 8i: Building Efficient Databases
> Publishers: Addison-Wesley
>
> Reviews at: http://www.jlcomp.demon.co.uk/book_rev.html
>
> Andrew Mobbs wrote in message ...
> >Jonathan Lewis <jonathan_at_jlcomp.demon.co.uk> wrote:
> >>
> >>This is the number of blocks that dbwr
> >>will attempt to find and write when called
> >>by a process that needs some buffers
> >>freed. It does not imply anything about
> >>continuous buffers, or single write sizes.
> >>
> >>In some versions of Oracle, it can be set
> >>by the value of _db_block_write_batch.
> >
> >I thought that "DBWR write chunk" was the number of writes that DBWR
> >would wait to return before putting the blocks from those writes back on
> >the LRU lists.
> >
> >It can be set (in 8.1.6 at least) by _db_writer_chunk_writes
> >
> >Andrew.
> >
> >--
> >Andrew Mobbs - http://www.chiark.greenend.org.uk/~andrewm/
  Received on Sun Feb 25 2001 - 06:38:26 CST