Re: row vs row.column level locking

On Sat, 17 Dec 2005 07:57:14 -0500, Serge Rielau wrote:

> The risks everyone takes by not running serializable/RR are real because
> most don't truly understand the implications of other isolation levels.
>
> Column level locking is safe, just too expensive.

Serge, I beg to differ. Sometimes, columns are logically related. It's easy to think of the logical disconnect among the two updating processes that would result in totally senseless claims. The world's most frequent last name is Wong. The world's most frequent first name is Mohammed. The conclusion that the world is full of people with the name Mohammed Wong is incorrect. That type of paradox is easy to construct and can easily break the logical consistency of your database. One would need to move all of the database mechanisms (foreign keys. trigers, constraints) from the row level to the column level to make is safe. I also believe that such "columns based database" would be too much for any computer of today, even things like HP SuperDome or SUn E15000 as every process would have to pass through an order of magnitude larger myriad of triggers, constraints and stored procedures just to update a single row.

PS:

---
My apologies to any Mohammed Wong who reads this. It wasn't personal.

-- 
http://www.mgogala.com