Archives of the TeradataForum
Message Posted: Tue, 29 Nov 2005 @ 15:06:10 GMT
| Subj: | | Re: What is LIBPM |
| |
| From: | | Geoffrey Rommel |
| | I've just started to get some exposure to Teradata - specifically MLOADS in a mainframe environment. The objective is to get some improved
performance both in elapsed time and CPU. | |
I don't think Strobe is going to help much. If performance is bad, the script probably needs a little tweaking. Three things come to mind.
1. Don't use too many sessions. Usually 16-32 is more than enough. You may want to run a few test jobs, loading the same data with different
numbers of sessions, to get an optimum figure. You can also code a single card saying 'MAXSESS=16' (and blanks in columns 73-80!) and assign it to
ddname MLOADCFG.
2. Don't take too many checkpoints. I suggest checkpointing no more often than every 4 blocks per AMP. For instance, if you're loading rows
that average 64 bytes in length, and your blocksize is 64K, you're getting about 1000 rows/block * 4 = 4000 rows; if you have 100 AMPs, your
checkpoint parameter should be at least 400000. If your ESCON/FICON connections are reliable, you can of course increase this figure.
3. Search the manual for "NULLIF Performance" to find a caution about the use of NULLIF in the FIELD statement. Generally, this sort of
processing is better performed in the database if possible.
Anybody else?....
| 
