Archives of the TeradataForum
Message Posted: Tue, 04 Jun 2013 @ 12:28:58 GMT
Hi -- there are so many possibilities. Some things to consider:
1) SHARING SPOOL -- are the overnight queries running under a single ID, or small number of IDs compared to daytime? If yes, then that could be the problem. All simultaneous queries running under a single ID share the spool on a FIFO basis.
2) TIMING OF STATS COLLECTION -- are stats collected at night when some tables are empty? Do these tables grow to a good size during the overnight ETL activity creating a stale stats scenario? Look at some of the staging databases and their tables.
3) STABLE ETL PROCESSES -- is the ETL cycle predictable? Do significant processing threads run late and/or long? Depending on what is happening this could be a subset of "1" or "2".
4) SKEW -- are any of the staging loads skewed? Any large tables with a skew number above 30? A few skewed tables or activities have an outsized impact on performance.
5) CHANGE -- a generic item = what is different about the night vs. the day (other than the sun has gone down)? Any recent changes in the nighttime activity set?
|Copyright 2016 - All Rights Reserved|
|Last Modified: 24 Jul 2020|