Re: Latch / Wait Events
From: vsevolod afanassiev <vsevolod.afanassiev_at_gmail.com>
Date: Sat, 14 Nov 2009 15:47:37 -0800 (PST)
Message-ID: <4406b4b3-8219-4f2e-a175-91bfd553aed9_at_y28g2000prd.googlegroups.com>
Date: Sat, 14 Nov 2009 15:47:37 -0800 (PST)
Message-ID: <4406b4b3-8219-4f2e-a175-91bfd553aed9_at_y28g2000prd.googlegroups.com>
- "rdbms ipc mesage" is idle event experienced by Oracle background processes (DBWR, SMON, PMON, etc) and it can be ignored.
- "log file parallel write" is event experienced by log writer. It is not experienced by application sessions.
Start from examining V$SESSION_EVENT/V$SESSTAT for sessions created by
application.
Let's say a session was connected 1000 seconds ago. It spent:
- 100 seconds on CPU
- 100 seconds on 'db file sequential read/db file scattered read"
- less than 10 seconds on other non-idle events (buffer busy/latch free, etc)
- idle rest of the time (SQL*Net message from client) This indicates healthy session
Let's say a session was connected 1000 seconds ago. It spent:
- 300 seconds waiting on 'latch free'
- 100 seconds on CPU
- 50 seconds on other non-idle events
- idle rest of the time This indicates latch issue.
In other words: start from comparing time spent on various wait events. Received on Sat Nov 14 2009 - 17:47:37 CST