Re: Flash / Data protection

From: K Gopalakrishnan <kaygopal_at_gmail.com>
Date: Fri, 2 May 2014 08:55:10 -0500
Message-ID: <CAN5iexECFBhD9LD-5fYYzniQPBsSMttzqp4=x83Lsockx6YS4w_at_mail.gmail.com>



George, Like all other answer it depends (on how the flash cache is configured.. Writeback/write through). Assuming you are on the latest (11.2.2.2.4 and above), the flash logging works similar to your description, except the ASYNC part. The writes (or commits) are deemed committed at least one write acknowledgement (usually from flash) is received. The rest of the writes go as usual. There is no pushing down to disk (like buffering).

-Gopal

On Fri, May 2, 2014 at 12:47 AM, George Leonard - Business Connexion < George.Leonard_at_bcx.co.za> wrote:

> It's friday, It's my curious day...
>
> Sure this is written some where.
>
> For Flash (F80 cards now) on Exadata storage servers connect to Exadata
> or SSC.
>
> Can someone confirm,
>
> For writes to a segment to write goes out to 2 or 3 storage servers
> (depending on the DG protection). All 2 or 3 accept the write (and
> acknowledges the write upstream as soon as the 1st has confirmed success),
> and then pushes ASYNC it down to the physical disk.
>
> Reads: only one copy of data is read, from disk, and buffered in the
> closest/in same storage server Flash or is all copies of the blocks read in
> the 2 or 3 storage servers and the blocks are buffered in all 3 storage
> server flash cards?
>
> Yours Sincerely
>
> ________________________________________
> *George Leonard*
> *Oracle Engineered System Specialist*
>
> Mobile: +27.82 655 2466
> eMail: george.leonard_at_bcx.co.za <george.leonard_at_oracle.com>
> Web: http://www.bcx.com <george.leonard_at_oracle.com>
>
>
>



--
http://www.freelists.org/webpage/oracle-l


image003.gif
Received on Fri May 02 2014 - 15:55:10 CEST

Original text of this message