Re: Flashback log file sync
Date: Sat, 12 May 2012 07:26:46 +0100
Message-ID: <0eydnYMgP434nzPSnZ2dnUVZ8hudnZ2d_at_bt.com>
"Mladen Gogala" <gogala.mladen_at_gmail.com> wrote in message
news:pan.2012.05.12.05.37.02_at_gmail.com...
| In the thread about LOB storage, Jonathan objected that I haven't really
|
| The event which caused the most time to be spent waiting was, as is to be
| expected, "direct path read".
Why do you expect to read when all you're doing is writing - (I think I know the answer to that question):
It would be interesting to see the session stats to see what the reads were, and how big the writes were.
I think the reads may have been "for flashback new" - to modify a block Oracle may have to read it and put it in the flashback log for writing before modifying it. In the case of very high volume the I/O for a LOB insert turns from "write" to "read / write (flashback) / write (lob)". (You may find that if you have a new tablespace the flashback activity is hugely reduced)
What's the full DDL for the table (use dbms_metadata.get_ddl) - are the LOBs nocache nologging, for example, and can this affect the flashback rate.
-- Regards Jonathan Lewis http://jonathanlewis.wordpress.com Oracle Core (Apress 2011) http://www.apress.com/9781430239543Received on Sat May 12 2012 - 01:26:46 CDT