FW: Strange Elapsed In Raw Trace

From: Cary Millsap <cary.millsap_at_hotsos.com>
Date: Wed, 13 Oct 2004 21:31:09 -0500
Message-ID: <004501c4b195$dd68e8a0$3e00000a@CVMLAP02>

Oracle tim values can wrap. But I would expect that seeing that event = would
be exceptionally rare.

The original posting looks like something Mike Brown of Glen Raven just discussed with me last week in Charlotte. If your db server is an NTP client, then it's possible for an e or ela value to look like a huge positive integer on platforms where the Oracle porters haven't been = careful
in regarding negative numbers as actual negative numbers. For example, consider the following sequence of events:

e0=3D gettimeofday returns n seconds since Epoch Oracle dbcall processing takes place
NTP adjustment takes place, setting the clock back k seconds More Oracle dbcall processing takes place e1 =3D gettimeofday returns n - k seconds since Epoch (k>0) Oracle calculates e =3D e1 - e0 =3D (n - k) - n =3D -k Oracle mistakenly emits e=3D-k in unsigned integer format ...which looks like a colossal positive e value

Happens at least on Oracle for HP-UX.

Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com
* Nullius in verba *

Upcoming events:

- Performance Diagnosis 101: 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_at_freelists.org =
[mailto:oracle-l-bounce_at_freelists.org]
On Behalf Of Larry Elkins
Sent: Wednesday, October 13, 2004 8:29 PM To: breitliw_at_centrexcc.com; oracle-l_at_freelists.org Subject: RE: Strange Elapsed In Raw Trace

No complaints ;-) They were happy with the *improvement*, or rather the user's ancestors were ;-)

An interesting thing is as the tracing continued I saw tim "wrap" back around:

EXEC
#5:c=3D27796875,e=3D807813015,p=3D89100,cr=3D262523,cu=3D44854,mis=3D0,r=3D= 33244,dep=3D1,og=3D
4,tim=3D18446744073387654595
*** 2004-10-13 17:34:19.679
EXEC
#5:c=3D27250000,e=3D471891123,p=3D98010,cr=3D262880,cu=3D34213,mis=3D0,r=3D= 29151,dep=3D1,og=3D
4,tim=3D471891123
*** 2004-10-13 17:45:43.523

I've been metalink'ing and google'ing (web sites and usenet) for more information about "tim" and it's derivation but haven't come up with anything yet. But I'm sure there's a rational explanation for this. Just trying to understand what it is. I've seen this before. And I've worked = in
DEC/VAX, MS-DOS (with Oracle 5), AIX, HP-UX, Solaris, and Windows environments. And I know for a fact that I've seen this before in this current Windows environment. Just trying to remember if I've seen it in other environments.

Regards,

Larry G. Elkins
elkinsl_at_flash.net
214.954.1781

> -----Original Message-----

> From: oracle-l-bounce_at_freelists.org
> [mailto:oracle-l-bounce_at_freelists.org]On Behalf Of Wolfgang Breitling
> Sent: Wednesday, October 13, 2004 7:56 PM
> To: oracle-l_at_freelists.org
> Subject: Re: Strange Elapsed In Raw Trace
>
>
> If I calculated that correctly, the 3rd exec took 584,942.4 years. Did
> anyone complain about performance?
>
> At 06:12 PM 10/13/2004, Larry Elkins wrote:
> >PARSE =

#5:c=3D0,e=3D3328,p=3D0,cr=3D0,cu=3D0,mis=3D1,r=3D0,dep=3D1,og=3D0,tim=3D= 3401329811

> >*** 2004-10-13 16:06:45.632
> >EXEC
> =
>#5:c=3D4562500,e=3D207030037,p=3D11799,cr=3D36578,cu=3D6669,mis=3D0,r=3D=

15004,d

> ep=3D1,og=3D4,t
> >im=3D3608386193
> >*** 2004-10-13 16:12:35.476
> >EXEC
> =
>#5:c=3D6484375,e=3D349792572,p=3D22658,cr=3D90642,cu=3D3842,mis=3D0,r=3D=

15801,d

> ep=3D1,og=3D4,t
> >im=3D3958240955
> >*** 2004-10-13 16:21:02.632
> >EXEC
> =
>#5:c=3D9546875,e=3D18446744065825247453,p=3D32226,cr=3D92572,cu=3D15058,=

mis

> =3D0,r=3D14816,
> >dep=3D1,og=3D4,tim=3D18446744069784377492
> >*** 2004-10-13 16:26:53.070
>
> Wolfgang Breitling
> Oracle7, 8, 8i, 9i OCP DBA
> Centrex Consulting Corporation
> http://www.centrexcc.com
>
> --
> http://www.freelists.org/webpage/oracle-l
>

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

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

Received on Wed Oct 13 2004 - 21:28:00 CDT