RE: OLTP compression slow update

From: Yong Huang <yong321_at_yahoo.com>
Date: Thu, 27 Dec 2012 13:20:13 -0800 (PST)
Message-ID: <1356643213.64060.YahooMailClassic_at_web184805.mail.gq1.yahoo.com>



> The reason being is that every session is contributing to the numbers
> -- like enqueues, latches, mutexes, or runtime.

Some but not all these metrics can be broken down into sessions. V$sesstat has session-specific 'enqueue waits'. You can also use event 10704 to trace a session's enqueue gets along the way. V$mutex_sleep_history has columns for requesting and blocking sessions, but you need to make sure these *historical* stats are really for the session you're interested in. V$sesstat has 'CPU used by this session' and 'CPU used when call started'.

Oracle never breaks latch statistics into sessions. You may frequently sample a specific server process with some tools for its calls to latch related functions, kslgetl etc. V$latchholder shows the sessions holding the latch right at the moment only. If there's latch wait, of course you'll have numbers in v$session(_wait) and related views for 'latch free' event which is session-specific.

Yong Huang

--
http://www.freelists.org/webpage/oracle-l
Received on Thu Dec 27 2012 - 22:20:13 CET

Original text of this message