Archives of the TeradataForum

Message Posted: Wed, 02 Jul 2003 @ 08:42:05 GMT

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

When statistics are not available, the optimizer dynamically does a "Random AMP Sampling" as & when required while processing a request. Based on the Table-ID, a random AMP is selected & statistics are constructed based on the subset of the table rows residing on that AMP only.

This method of generating statistics is very much dependent on the evenness of distribution of the table rows across the AMPs. If the AMP sampled doesn't have the correct share of the table, the statistics thus constructed would definitely be skewed & might mislead the optimizer into generating sub-optimal plans.

On the other hand, if the distribution is even, this approach would be better since the optimizer would have the most current statistics.

Sanjay

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