[Dev] Re: Data outside the box

Paul Snively psnively at mac.com
Sat Nov 23 21:46:53 PST 2002

Hash: SHA1

On Saturday, November 23, 2002, at 08:15  PM, patrickdlogan at attbi.com 

> Here's what I'm thinking: what's the difference between a table and an
> index? Well, and index is just for one column typically. So what's the
> difference between a table and a set of indexes, one per column?
> One answer is: tables are not conducive to working on columns (most
> common), so you need the indexes anyway. So if you've got the data in
> indexes, what do you need the table for?

> I'm asserting there are indexes but not tables! Columns (indexes) are
> coming and going, or portions of them. Tables can and should be wide,
> but you don't need to bring an entire row into memory, just the
> columns you need. That's essentially what you do with indexes, so why
> store data in row format at all when most of the time you want to work
> with columns.
> What if data were organized by columns rather than rows? They would in
> essence *be* indexes.

> What if you could add and remove columns in the model without
> affecting the organization of the rest of the table?
I hate to sound like a broken record, but all of this sounds an awful 
lot like <http://www.equi4.com/metakit/format.html>. The only issue 
with MetaKit would seem to be the current implementation's wanting to 
mmap() the entire file into the address space.

> -Patrick
Best regards,

Version: GnuPG v1.0.7 (Darwin)


More information about the Dev mailing list