Re: Hot index block split on a very busy table--what is the impact


Disregard as much of my previous post as you think fit.
My brain has just woken up to the latches being CBC,
not shared pool and library cache.

The branch block split may still be relevant - and the increase
in processes cause by the split; but you don't seem to have
a consequential shared pool problem.

Regards

Jonathan Lewis
http://www.oracle.com/technology/community/oracle_ace/ace1.html#lewis

The Co-operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/faq/ind_faq.html

Cost Based Oracle: Fundamentals
http://www.jlcomp.demon.co.uk/cbo_book/ind_book.html


----- Original Message ----- From: "Zhu,Chao" <zhuchao@xxxxxxxxx>
To: <oracle-l@xxxxxxxxxxxxx>
Sent: Wednesday, August 30, 2006 2:53 PM
Subject: Hot index block split on a very busy table--what is the impact


hi, all,
   We have a few database with very high concurrent access busy tables,
that some indexes of the busy tables could be accessed 1.5K-4K times per
second.  (PK lookup)
   I am wondering , when such index got block split (at the root, or branch
level), what will be the impact on the system.
   Index block split is said to be an expensive operation, during the block
split on branch/root block, what the other sessions that is doing select
based on this index, be waiting on? There is a wait event named:  Index
Block Split with p1,p2,p3 pointing to block address, level.
  Document "Description of oracle 7 Wait event and enqueue" says it will
yield CPU.
  We have a few production incident when load suddenly jump from 10, to
several hundred, and during that time nothing can be done. From our
monitoring tools, it is mainly "latch free" contention inside oracle, and it
is cache buffer chains. The load spike matches the oracle description of
block split, but the wait event does not match.  And because the load spike
time is very short, we can't capture the actual wait event/p1,p2,p3 during
the exact time of the load spike time.
 Anyone have similar problem/insight into this issue?
 Is there some v$ view to track, (v$segstat does not) have log about which
index go through how many time block split?  Or any other effecitve way?


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


Other related posts: