Archives of the TeradataForum
Message Posted: Mon, 03 Mar 2008 @ 14:11:03 GMT
| Subj: | | Re: MLOAD application phase |
| |
| From: | | Victor Sokovin |
| | Thank you very much for your suggestions. It worked like a charm. We simply partitioned the table on a monthly basis and it loaded in a
few minutes. | |
This sounds as if the load takes place into an empty table, which is probably the case if the bulk of the load was done in the March
partition.
This experiment seems to confirm that the speed of the original job was negatively influenced by the presence of the data in the table. My
guess was that the time was probably spent on duplication checks (new data being checked against the current data) but that would only account for
a couple of minutes out of 30+ min spent on the entire job (according to the OP).
What other influence could the existing data in the table have on ML? Would ML attempt some kind of data redistribution in the existing blocks?
We should not forget that the table has a UPI on it so we would think that there was not that much to do for ML. I am still thinking about this
but so far I don't see any valid ML activity which would "explain" the need to slow things down.
Any ideas on this?
Victor
| 
