Database File Systems

Back in January, Microsoft announced that their next version of Windows would consolidate their multifarious data-stores into one package. Currently, they have NTFS, SQL-Server and the Exchange format to house and index data. This was reported and covered in detail by the Register in XP successor Longhorn goes SQL….

Recently, the Register ran an interview with Benoit Schillings and Dominic Giampaolo, both of BeOS fame, called Windows on a database…. The BeOS probably came the closest to offering a database underlying a file system. As the interview reveals, in fact, the original version, was a database:

“Benoit Schillings was one of Be Inc’s first employees, and authored the original user space database server. This was later superseded by a more conventional approach: BFS, a fast, 64bit journaled file system written by Dominic Giampaolo, which had many database-like properties.”

The BeOS had arbitrary attributes that you could set on any node in the file system, so that a node in the file system didn’t even have to have data, per se, but could be entirely described by it’s attributes. This is very powerful for the user when combined with the instantaneous updating the BeOS provided. The native BeOS email client stored emails in the file system this way, which allowed users to use file-system queries to find emails and any email client could access the same database of information. Dominic mentioned an interesting use:

“I had a little window for my fiancée, and whenever she’d send me email it would pop up. Doug Fulton, actually, he would put everything in a big flat directory and use queries to find the stuff. … (Register) Oh I do that with my MP3s on BeOS. There are subdirectories there but I can’t even remember what they are”

Imagine not having to impose a hierarchy on your data, but rather use queries to extract the information you want. The power of the system is directly proportional to the amount of detail you ascribe to your database entries.

Name Spaces As Tools… is a longish paper on the same topic, addressing the trap we’ve fallen into with the one-dimenional view we have of data. In the current system, the user is forced to categorize data, and usually along just one dimensions. A hierarchical tree does not allow storing a file in both the ‘Male’ folder under ‘Gender’ and in the ‘Brown’ folder under ‘Eye-Color’. Almost everyone has had to make these decisions before. The article goes into considerable technical detail in trying to find a solution.

A rich structure of detail through which dimensionality is extracted rather than imposed. That sounds a lot like Ted Nelson’s Xanadu, of which the ZigZag project is the latest outgrowth. I highly recommend the Tutorial if you’re looking for a better way of storing, navigating and organizing information than hierarchical folders.