Archives of the TeradataForum

Message Posted: Fri, 27 Apr 2007 @ 13:38:38 GMT

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

Some findings:

1) MV analysis may be done in a single pass using the group by grouping sets

BUT it is HORRIBLY SLOW.

It is much faster to collect the stats against the individual columns.

Testing conditions:

1) Small test box, 1 node, 7 amps

2) 26.8M rows; 5gb of total space, skew is negligible

Testing Constraint

1) Only return values that are present on at least 1% of the rows

Testing outcomes

1) Analysis time using grouping sets = 60 min;

2) Analysis time collecting column by column < 15 min

Conclusion

1) It is much faster to collect the stats against the individual columns.

Next Step

1) Try to use external programs featuring arrays assigned to each attribute and perform the analysis in a single pass -- starts to get complex, may still be slow.

2) Investigate the layout of index and colstats fields.

Rgrds,

Bill Gregg

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