All Forums General
rajesharra 8 posts Joined 03/12
08 Mar 2013
AMP - Storing the Record

Hi There,

If i  have created nupi on the sex column, definatly it has skew ness. Think those two amps are filled (while others are free), we got one more record in to the table. what will happen?

My assumption is the record will not insert, as per the rowid/rowhash.
Am i correct?

dnoeth 4628 posts Joined 11/04
08 Mar 2013

Short answer: Yes :-)
If you try to insert a row which hashes to the AMP which is running out of perm for this database you'll get a "No more room in database" error, i.e. 'M' or 'F' will fail, but 'X' might work.


rajesharra 8 posts Joined 03/12
11 Mar 2013

Thank you

rajesharra 8 posts Joined 03/12
11 Mar 2013

Hi Dnoeth,
When i said the same to my friends, they are saying if this is the case (gives error no more room), then it is a bug of TD.
They are saying there should be some mechonisam which is introduced by the Tearadata team, to handle this situation,because the space is available in other AMPs.
Can you please explain more
Rajesh A

Adeel Chaudhry 773 posts Joined 04/08
11 Mar 2013

In such a scenario .... one should think before choosing a PI.
It is not a bug of Teradata .... it just doesn't expect a PI to be a column which have only 2 unique values .... imagine a warehouse for some Telecom and putting PI on such combination would obviously doesn't make sense. Teradata is built for DHW .... where the data is dynamic, changing and growing and PIs are not made on such columns.
Selecting correct PI is also important to completely utilize the true parallel nature of Teradata. Which you are not going to use in such a case. Every query hitting the table will only be utilizing 2 AMPs .... and usually Teradata Nodes does have alot more AMPs.
To the friends, well its the way Teradata stores data .... and if it is given a false parameter to store data .... its our friend's bug .... :).
As a general rule, PI should be the column(s) which are selected on the basis of table's usage and provide best data distribution among 'ALL' AMPs with minimum skewness.

-- If you are stuck at something .... consider it an opportunity to think anew.

You must sign in to leave a comment.