 |
 |
Archives of the TeradataForum
Message Posted: Thu, 17 Aug 2006 @ 10:26:01 GMT
| Subj: | | Re: Block size for Tactical Queries |
| |
| From: | | John Graas |
TeradataForum wrote:
| | If a particular table services both Decision Support (involving table scans) and Tactical Queries (hopefully not involving table scans),
how best to design the block size ? | |
| | (Even though I seem to remember Stephen Brobst touching on this point at the Salzburg Tech conference, I can't for the life of me remember
the detail.) | |
| | Is it best to have a small block size and rely on the automatic Table Scan 'scooping up' multiple blocks at one time approach, or should I
rely on the fact that a non-Table Scan read would not even read the whole block size ? | |
I believe the consensus is that, if a table has mixed decision support and tactical query access, that the maximum block size (e.g. 64K, 128K)
is best.
The cost of the extra I/Os for the DS queries is going to be greater than the cost of extra data retrieved for an I/O for tactical queries.
Also -- IIUC -- the extra I/Os caused by DS queries accessing small blocks could actually be detrimental to concurrent tactical queries in a heavy
mixed workload environment.
jdg
www.jgraas.com
| |
