Comments on: How parallel http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/ Just another Oracle weblog Sat, 18 May 2013 11:04:10 +0000 hourly 1 http://wordpress.com/ By: Plan Notes « Oracle Scratchpad http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-39049 Wed, 05 Jan 2011 08:46:36 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-39049 [...] but it seems likely that they’re intended to supply the answer to the common question “How parallel did my query go, and why [...]

]]> By: Jonathan Lewis http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-30854 Wed, 16 Apr 2008 18:58:04 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-30854 Dion Cho,

That’s a useful little paper. Be careful though, I just tried “high, all” on a little query and ended up with a 200 Mb trace file before I killed the query ;(

]]> By: Dion Cho http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-30853 Wed, 16 Apr 2008 09:02:23 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-30853 “_px_trace” might be the most convenient way.

Metalink doc# 444164.1

]]> By: Jonathan Lewis http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-28688 Thu, 10 Jan 2008 18:48:42 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-28688 I had a little discussion with Greg Rahn recently on the topic of DFOs (data flow operations), and he pointed out that the terminology had changed, although some of the column names and statistics names had not.

In the output from v$pq_tqstat above, there is a column labelled dfo_number – and I had been using the term DFO to refer to all the table queues (TQ_ID) within a single DFO. Greg refers to these as parallelizers, and I think it’s the thing I call a table queue that is now called a DFO.

However, if you want to see a plan with multiple DFO sections (as per v$pq_tqstat) in it, then a parallel query that manages to serialize with some non-mergeable views is one way of doing it. For example:


create table t1 as
select
       rownum id, object_name, object_type, owner
from
       all_objects
where
       rownum <= 10000
;

create table t2 as
select * from t1;

create table t3
partition by hash (id) partitions 8
as
select * from t1
where rownum <= 1;

begin
       dbms_stats.gather_table_stats(
               user,
               't1',
               cascade => true,
               estimate_percent => null,
               method_opt => 'for all columns size 1'
       );

       dbms_stats.gather_table_stats(
               user,
               't2',
               cascade => true,
               estimate_percent => null,
               method_opt => 'for all columns size 1'
       );
end;
/

alter session enable parallel dml;

spool pq_multi

explain plan for
select
       /*+
               parallel(t1,7)
       */
       rownum, t1.object_name, t1.object_type, t1.owner
from
       t1,        
       (
       select
               /*+
                       parallel(t1,5)
               */
               rownum, t1.object_name, t1.object_type, t1.owner
       from
               t1,        
               (
               select
                       /*+
                               ordered
                               parallel(t1,3)
                               parallel(t2,3)
                               use_hash(t2)
                       */
                       rownum,
                       t1.object_name,
                       t2.object_type,
                       t2.owner
               from
                       t1, t2
               where
                       t2.object_name = t1.object_name
               and        t2.object_type = t1.object_type
               and        t2.owner = t1.owner
               )        v1
       where
               v1.object_name = t1.object_name
       and        v1.object_type = t1.object_type
       and        v1.owner = t1.owner
       )        v2
where
       v2.object_name = t1.object_name
and        v2.object_type = t1.object_type
and        v2.owner = t1.owner
;

]]> By: kaiyaohuang http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-28661 Thu, 10 Jan 2008 13:36:23 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-28661 Hi, I think that here DFO means DFO tree here, could you give me an example of query that using multiple DFO trees?
It’s best if you can also mail to kaiyao.huang@gmail.com:)

]]> By: David Aldridge http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-3964 Wed, 14 Mar 2007 14:50:02 +0000 http://jonathanlewis.wordpress.com/2007/03/14/how-parallel/#comment-3964 One advantage that we users of PQ have at least is that the overhead of all the optional tracing is relatively small, so automaticaly dumping post-query wait stats and plans, or pre-emptively turning on tracing, is not some system-halting issue.

I like this 10391 — useful stuff, and thanks.

]]>