![](//teradataforum.com/imgs/ptdfw1.gif) |
![](//teradataforum.com/imgs/ptdfw2.gif) |
Archives of the TeradataForum
Message Posted: Thu, 07 Jun 2007 @ 13:33:57 GMT
Subj: | | Re: Designing wide (fat) tables |
|
From: | | Victor Sokovin |
| Normally these large tables are also designed with compression on some of the columns in mind. The bummer being that once you have too many
columns, your table header increases in size, and then the number of values you can acutally compress reduces (as they need to be there in the
table header, which is of fixed size). I am not sure on the size of table header and whether it has improved in later releases, but yeah it used
to be a common occurrence in the past to see a table header overflow with too many cols+compression+indexes ...... So that would be one silent
beast I would recommend you to watch for. (sometimes the pure motivation on going for a flat table is to avoid joins and make use of compression
in storing those extra columns, a noble desire) | |
Fair warning, Joseph. Table header size increased in recent releases but the number of factors contributing to the overflow also increased.
Take, for example, the relatively new and tempting partitioning options. But also the good old VARCHAR columns should not be forgotten as they
also contribute to the header.
I recall and can find the thread where some examples and calculations related to this topic can be found:
www.teradataforum.com/teradata/20050505_140933.htm
I can't add much more. I understand that wide tables might be unavoidable at times and they do the job they are designed to do but they are not
my preferred way of storing data.
Regards,
Victor
| |