Thursday, April 27, 2006

(RDBMS) v (¬RDBMS)

My previous post was in response to Keith's writing regarding the possibility of escaping object relational mapping problems by avoiding altogether RDBMS's. The discussion of where and how relational databases are to be used turned out to be a hot topic on the blogosphere* these days, mostly fueled by Tim O'Reilly's nine-part saga on data management in Web 2.0* companies. The three most salient points I got from Tim's posts were: there is a preference for clustering by data partitioning instead of general replication strategies; predictably there also is a preference for open source relational db systems (MySQL being the front-runner); and there is a belief that read-mostly data is better served from flat files (sometimes with sophisticated replication mechanisms) than from RDBMS'. There is a lot of interesting info there, I recommend reading the whole thing.

Speaking of read-mostly data, I found this cool project via Joe Gregorio. It's a relational database system being built at MIT optimzed for this kind of data. Another interesting system for dealing with simple read-mostly data is Google's BigTable, that seeks to achiveve high throughput by adopting of a very simple data model, whithout integrity constraints nor transactional capabilities, structured in a highly distributed and fault-tolerant fashion (as would be expected of google). It also features native data versioning support.

PS: Yes, this post was written in english. At least something approaching what one might call "english"... The short motive is that I just felt like it. The longer one is that I need to practice my english at least as much as I need to practice my portuguese, and writing practice is one of the primary reasons for this blog's existence (yes, I'm a selfish bastard).

* buzzword count: 2. I'll try to do better next time.

No comments: