Comments on: Analytic Agony

By: Histogram Generation « Oracle Scratchpad

Histogram Generation « Oracle Scratchpad — Thu, 13 Sep 2012 17:21:39 +0000

[...] it; it’s a window sort of the entire sampled data set which could be quite large and, as we’ve seen elsewhere, could be very [...]

By: Jonathan Lewis

Jonathan Lewis — Mon, 01 Aug 2011 11:03:03 +0000

I can't think of any definite reason why not - as far as I know there is nothing that forces analytic functions to run serially. I suggest you post the query and the execution plan to the OTN database forum and ask for ideas.

By: Sudheer

Sudheer — Tue, 26 Jul 2011 12:47:11 +0000

I have noticed WINDOW (SORT) operation is running serial (I mean only one slave is processing the data) , do you have any suggestions to improve this.

By: Dom Brooks

Dom Brooks — Wed, 09 Mar 2011 17:15:46 +0000

Jonathan,

> it would be nice if the wait states reported the line of the execution plan that they were operating on when they started to wait

I know it’s an old thread and comment but just a couple of thoughts on this.

As you will know, ASH gives the ability to do some of that. Using the v$active_session_history buffer, I know it’s only sampled data every second but

1. it reports waits with execution lines via SQL_PLAN_LINE_ID / SQL_PLAN_OPERATION / SQL_PLAN_OPTIONS and EVENT.

2. real-time SQL Monitoring can report similar (also based on the ASH data but also tying in with longops data which is useful when for operation progress % when using it in real-time which obviously it doesn’t have to be despite the title).

Cheers,
Dominic

By: Jonathan Lewis

Jonathan Lewis — Tue, 16 Feb 2010 18:55:52 +0000

Alex.
That sounds like a nice idea – but there are lots of little changes like that that would be really nice so it’s hard to make a very strong case for any one of them. For example, it would be nice if the wait states reported the line of the execution plan that they were operating on when they started to wait.

Since trace files tend to be after the event, linking a trace file with a dynamic performance view might be a little awkward for the DBA.

By: Alex Haralampiev

Alex Haralampiev — Tue, 16 Feb 2010 04:35:58 +0000

Hi Jonathan,

I think we can correlate the “one-block-at-a-time” reads with phases when the v$session_longops.OPNAME=’Sort Output’ and V$SESSION_WAIT.event=’direct path read temp’; The bad news is currently there is no way to inject the v$session_longops.OPNAME values in the 10046 traces so we can’t get more objective breakdown. The good news is I managed to convince a few people at Oracle (Carlos Sierra and Hector Pujol) to open a new enhancement request #9381412 for the following:
“For diagnostic purposes it helps to correlate the value
v$session_longops.opname with waits or CPU related to the OPName. This
request is for additional info to be placed into the 10046 trace file that
would allow TKProf or other tools to report how much cpu, waits, and elapsed
time occured for an OPName as well as track specific waits that occurred
under that OPName. To illustrate, this would appear as:

LONGOP #2 opname=”Sort Output” cpu=12345 elapsed=4567
WAIT #18: nam=’direct path read temp’ ela= 132 tim=1232507925946426
WAIT #18: nam=’direct path read temp’ ela= 141 tim=1232507925947219
WAIT #18: nam=’direct path read temp’ ela= 132 tim=1232507925947762

TKprof could then report how much time occured under each longop.opname ”

If you think that this is a good idea please let me know if/how we can make a stronger case for such enhancement.

Rgds,
Alex

By: Jonathan Lewis

Jonathan Lewis — Sun, 07 Feb 2010 17:24:09 +0000

Greg

Thanks for the update.

By: Greg Rahn

Greg Rahn — Sun, 07 Feb 2010 16:27:10 +0000

This related to bug 9041800 and there is a 10.2.0.4 backport available as of 01/29/10. For other versions/platforms, request a backport from Oracle Support.

By: Log Switching « Oracle Scratchpad

Log Switching « Oracle Scratchpad — Fri, 01 Jan 2010 15:08:08 +0000

[...] of uses, and the lag/lead functions are among my favourites. I always like to remind people of the sorting overheads involved with analytic functions – but if you are prepared to accept the overhead, the [...]

By: Jonathan Lewis

Jonathan Lewis — Sat, 17 Oct 2009 10:33:24 +0000

Henish,

The book is based on 9.2 with some comments on differences between 8i and 10g.

The details of hard and soft limits based on the pga_aggregate_target changed quite siginficantly across versions. The most comprehensive description of 10g behaviour comes from a paper by Joze Senegacnik