I think this is just a case of less than clear documentation. I looked at
the 9.2 beta doc and that entire phrase has disappeared, which sort of
reinforces my opinion.
What happens really is what I sent out in mid January and what I told Nuno
to go looking for. If you need that sent again I can probably find the
email, but a search of the newsgroup should locate it. The pinning, logging
etc. is normal processing we go through regardless of whether this is RAC or
not.
--
HTH. Additions and corrections welcome.
Pete
Author of "Oracle8i: Architecture and Administration Exam Cram"
"Controlling developers is like herding cats."
Kevin Loney, Oracle DBA Handbook
"Oh no, it's not. It's much harder than that!"
Bruce Pihlamae, long-term Oracle DBA
"Slava" <leichivp_at_my-deja.com> wrote in message
news:a775741a.0201312316.73eeab73_at_posting.google.com...
> "Pete Sharman" <peter.sharman_at_oracle.com> wrote in message
news:<HRd68.6$8X1.93_at_inet-nntp1.oracle.com>...
> > I may not be understanding your issue correctly here, but the log buffer
> > flush is not tied to the block transfer through the interconnect AFAIK.
The
> > log buffer flush is an independent event.
> >
>
> Well,
> this is citation from Oracle9i Real Application Clusters Deployment
> and Performance Guide, chapter 6:
> "
> Elimination of I/O for Forced Disk Writes of Blocks
> Cache Fusion practically eliminates disk I/O for data and undo segment
> blocks by transmitting current block mode versions and consistent-read
> blocks directly from one instance's buffer cache to another. This can
> reduce the latency required to resolve writer/writer and reader/writer
> conflicts by as much as 90 percent.
> Cache Fusion resolves concurrency, as mentioned, without disk I/O.
> Cache Fusion expends only one tenth of the processing effort that was
> required by disk-based parallel cache management. To do this, Cache
> Fusion only incurs overhead for:
>
> Pinning a current block, logging the changes to the block, forcing a
> log flush, and sending the block
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> Or when requesting a consistent read version:
>
> Processing the request and constructing a consistent-read copy of the
> requested block in memory and transferring it to the requesting
> instance
>
> On some platforms this can take less than one millisecond. "
>
> Could you clarify this one ?
>
> Regards,
> Slava.
Received on Mon Feb 04 2002 - 13:48:00 CST
