RE: Missing time in 10046 trace file
- From: "Cary Millsap" <cary.millsap@xxxxxxxxxx>
- To: "'Oracle-L (E-mail)'" <oracle-l@xxxxxxxxxxxxx>
- Date: Fri, 1 Oct 2004 11:46:18 -0500
This is why I so hate the term "wait"...
The ela value of a 'latch free' timed event is the duration of a =
sleep-like
OS call. You're referring to the quantity D in the following relation:
D =3D e[dbcall] - (c[dbcall] + sum(ela[dbcall]))
The large D value that you describe is probably the result of process
preemption (as John has suggested).
An Oracle kernel process consumes a lot of CPU before it ever gets to =
the
point at which it will execute the 'latch free' timed event. If a lot of
kernel processes are fighting for the same small set of latches, then =
there
will be a lot of CPU capacity wasted, and in severe cases you'll see =
lots of
time spent preempted. ...Which is, I believe, exactly what you're =
seeing.
Read "Why you should focus on LIOs instead of PIOs" at www.hotsos.com =
for
the gory details.
Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com
* Nullius in verba *
Upcoming events:
- Performance Diagnosis 101: 10/5 Charlotte, 10/26 Toronto, 1/4 Calgary
- SQL Optimization 101: 10/18 New Orleans, 11/8 Dallas, 12/13 Atlanta
- Hotsos Symposium 2005: March 6-10 Dallas
- Visit www.hotsos.com for schedule details...
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx =
[mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of John Kanagaraj
Sent: Thursday, September 30, 2004 5:54 PM
To: 'pfitzger_au@xxxxxxxxx'; Oracle-L
Subject: RE: Missing time in 10046 trace file
Paul,=20
>tim value of the FETCH 2276365740 from the time value of the EXEC
>2276356674 I get approx 90 seconds (9066 centi-secs) which is nowhere
>close to adding all the ela's of the waits (48 centi-secs). Am I
>looking at what they call unmeasured/unaccounted-for time?
Note that latches can 'sleep' and while they are sleeping, this time =
isn't
counted under 'waiting'. This sleep is successivley for longer =
periods... I
think Steve Adams discussed this very clearly in his book. If you look =
at
Statspack snapshots between these periods, you might see more than usual
values under the SLEEP[n] columns. And yes, when CPU thrashing occurs, =
the
unaccounted time increases as the context switching times are not =
counted
within Oracle....
Hth,
John Kanagaraj <><
DB Soft Inc
Phone: 408-970-7002 (W)
Listen to great, commercial-free christian music 24x7x365 at
http://www.klove.com
** The opinions and facts contained in this message are entirely mine =
and do
not reflect those of my employer or customers **
--
//www.freelists.org/webpage/oracle-l
--
//www.freelists.org/webpage/oracle-l
Other related posts:
