Granted, I don’t understand the inner-workings of the column-based databases, but if they rely on the mere fact that data is stored together, then I am doubtful that the solution would really work all that well. The Oracle functionality listed by many of the posters here–bitmaps indexes especially–actually allow the database to retrieve the row quickly no matter where or how it’s stored, and that is the paradigm that seems most important.

is the column-orientation-paradigm another name for vertical partitioning in traditional RDBMS?

It’s certainly an interesting subject. Thinking about it, if all the data for a particular column was placed next to each other on disk, rather than spread over many blocks as part of whole rows of (unrelated) data, then logically it should be faster to retrieve. I’m conscious though that other things come into play, IOTs or bitmap indexes can do much the same thing, and there’s all other factors involved including how scalable the servers are, how much duplication of data has to go on if loading into a column-based database, and so on.

That said though, it’s definately interesting and something I wouldn’t mind raising (informally, if appropriate) at the upcoming BIWA conference, where a lot of the Server Tech PM guys will be attending. Watch this space, as they say.

Of course I have no idea how this would work with real data and the need to update attributes when people move house or dye their hair, or how it would work with non-key searches such the people with ‘I’ as the second letter in their name.

It could be a “playtime” thing to look at

It is difficult to see how storing in (potentially as I don’t know the column database design) an aggregate fashion that you could drill down further and back again!

IOTs or manual ORDER BY approaches can achieve this, although they are limited to a specific column or set of columns to colocate by.

Colocation, as I’ve mentioned once or twice on my blog, is useful in enhancing the degree of compression achieved.

Bitmap and Bitmap Join Indexes can provide orders of magnitude improvement on column based queries like “Brown Hair” and “Socio Economic Group G”.

It’s funny, I was going through my blog aggregator and read the post by Michael Stonebraker before yours and I thought to myself…lots of “marketing speak there” but no actual substance or, dare I say it, benchmarks…then I began to think, I wonder who he works for…then I read your article.

I’m not saying there isn’t a market for the column based approach but it does seem like a niche thing to my mind…as you say, it would be nice to do some benchmarking and testing of it’s approach in comparison to Oracle (or other RDBMS).

I don’t particularly regard Oracle as a “one size fits all” – I regard it as a multi talented offering, which, with the appropriate people using it (to take advantage of the most appropriate features), can achieve excellent results in many disciplines…that’s something I did agree with the Stonebraker article on…it’s the people that make the difference, no matter how clever the product.