I propose enhanced way to split the table into multiple chunks by ROWID

Question

There is a good methodology, called as ROWID Parallelism, introduced by Thomas Kyte ([http://asktom.oracle.com/pls/asktom/f?p=100:11:0::NO::P11_QUESTION_ID:10498431232211]) and explained in his book "Effective Oracle by Design". But there is a one problem with base query – max number of chunks is constrained by number of extents, allocated for table. So max number of parallel processes is constrained too. To overcome this problem I propose new query based on blocks rather than extents. In this query max number of chunks is constrained only by number of blocks, allocated for table, rather than allocated extents:
define TNAME=<write table name here>
define CHUNKS=<write desired number of chunks here>
with extent_blocks as
( select relative_fno,
block_id,
blocks,
( sum( blocks ) over( order by relative_fno,
block_id rows unbounded preceding )) - blocks + 1 as block_1,
sum( blocks ) over( order by relative_fno,
block_id rows unbounded preceding ) as block_2
from dba_extents
where segment_name = UPPER( '&TNAME' )
and owner = user ),
range_blocks as
( select row_num,
b1,
b2
from ( select ROWNUM as row_num,
( part_n *( ROWNUM - 1 )) + 1 as b1,
case
when part_n * ROWNUM > s.last_n
then s.last_n
else part_n * ROWNUM
end as b2
from ( select ROWNUM
from DUAL
start with dummy = dummy
connect by dummy = dummy ),
( select sum( de.blocks ) as last_n,
CEIL( sum( de.blocks ) / &CHUNKS ) as part_n
from extent_blocks de ) s
where ROWNUM <( &CHUNKS + 1 ))
where b1 <= b2 )
select r1.row_num n,
DBMS_ROWID.rowid_create( 1,
obj.data_object_id,
r1.relative_fno,
r1.block_id +( r1.b1 - r1.block_1 ),
0
) min_rowid,
DBMS_ROWID.rowid_create( 1,
obj.data_object_id,
r2.relative_fno,
r2.block_id +( r2.b2 - r2.block_1 ),
32767
) max_rowid
from ( select rb.row_num,
rb.b1,
eb.relative_fno,
eb.block_id,
eb.block_1
from range_blocks rb,
extent_blocks eb
where rb.b1 between eb.block_1 and eb.block_2 ) r1,
( select rb.row_num,
rb.b2,
eb.relative_fno,
eb.block_id,
eb.block_1
from range_blocks rb,
extent_blocks eb
where rb.b2 between eb.block_1 and eb.block_2 ) r2,
( select data_object_id
from user_objects
where object_name = UPPER( '&TNAME' )) obj
where r1.row_num = r2.row_num
Denis Komarov

Answer 1

All well and good, but a couple of questions for you.
1) any specific reason why you'd post that here as opposed to the site you've linked to?
2) "So max number of parallel processes is constrained too."
This is true of your method as well, seeing as how you can only have so many parallel processes before you stress the machine to the point of having the parallel processes fight each other for resources. I wouldn't imagine you can actually make use of the number of parallel processes you propose (you seem to want a very high number of them).
In practicality, i think the method demonstrated by Tom would be more than sufficient, but please do feel free to put together a test case and prove me wrong......