> This model is completely removed from how the data is physically > stored, eg whether in rows first or in columns first, and the > physical store is determined just by the DBMS behind the scenes, and > hence is an implementation detail. The DBMS can arrange how it likes > in order to satisfy the logical model in a way that performs well. > The users do not contort their schemas beyond what is logically clean > in order to gain performance; having a clean schema will let the DBMS > infer this automatically.
This may be true in theory, but if it were really true in practice, why would anyone bother building databases with different storage models? How a database does what it does impacts everything from where you should use it to how you design your schema to how you write your queries. And if you think that no one contorts their schemas to make their queries run faster then, umm... Well, I don't think I need to say anything else about that. > According to some comments, Vertica (a column-store maker) is making > the same case that I am, which is just to have a logical clean > schema, and performance benefits will automatically follow from that. They say that because their system performs better when your schema is simpler. It's not just a philosophical belief that simpler is better. > I will also note that a column-based store essentially works like a > heavily indexed row-based store, in which there is an index on every > key or every column, and so all searches, which includes those on > which joins are performed, can/do look in what is otherwise indexes. Conceptually I agree, though I do not think you could build an equally-effective system by indexing a row store. > This is potentially slower for updates (or maybe not), but can be > faster for queries, depending on circumstances. But... But... How the data is stored is an implementation detail that users don't need to worry about, right? So who cares if updates are slower? The logical model is the same, right? Thank you for making my point. :) -T ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------
