Archives of the TeradataForum

Message Posted: Fri, 02 Jan 2004 @ 16:15:37 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

McBride, Michael wrote:

It depends...

A year ago i run some tests on a 4-node 5300 (2GB memory, 36 AMPs � 36519 cylinders, 144 disks 36GB RAID 1). The following query was similar to the requested one:

table TPCD.lineitem
13142 datablocks, 468 cylinders
6.004.101 rows
tablesize: 905.870.336 Byte

/***
   select all data for rows with the highest quantity for
   each partkey/shipdate combination -> 237.288 rows
***/

     sel * from
       lineitem
     where
       (l_partkey, l_shipdate, l_quantity)
       in
         (
           sel
             l_partkey, l_shipdate, max(l_quantity)
           from lineitem l1
           group by 1,2
         );

-- cputime: 124,58
-- IOs: 152616


     sel
       *
     from lineitem l1
     qualify
       rank() over (partition by l_partkey, l_shipdate
                    order  by l_quantity desc) = 1;


-- cputime: 96,87
-- IOs: 87376

Especially if you have to use nested subqueries or it's a joined view instead of a base table OLAP functions are usually more efficient.

Dieter

	<Prev	Next>		<<First	<Prev	Next>	Last>>