Skip the navigation

The Road to Cairo

By Nicholas Petreley
April 8, 2002 12:00 PM ET

Computerworld -





Join the online discussion about this column.




The year is 1992. Jim Allchin reveals Microsoft's plans to deliver a version of Windows NT, code-named Cairo, in 1994. Cairo is slated to use an Object File Store (OFS) as its file system. OFS is an object-oriented database designed to make it easy to search documents and other structured data by content. Fast forward to 2002. The continued status of OFS can best be described by the famous Monty Python cheese shop sketch (see www.montypython.net/scripts/cheese.php for the complete script). In the sketch, the customer (John Cleese) asks the clerk (Michael Palin) for every conceivable type of cheese, but the shop is out of stock on every item. This leads to the now classic exchange, which exploits Cleese's exceptional timing and delivery:
Cleese: "It's not much of a cheese shop, is it?"
Palin: "Finest in the district, sir!"
Cleese: "Explain the logic underlying that conclusion, please."
Palin: "Well, it's so clean, sir!"
Cleese: "It's certainly uncontaminated by cheese."
Similarly, 10 years after Allchin's initial promise, Windows remains uncontaminated by many of the features originally slated for Windows NT and Cairo, including OFS.
There is a very simple explanation for the delay in the case of OFS. Microsoft thought it might need OFS to win the battle against OS/2, which already had an object-oriented foundation and threatened to include a database-oriented file system. As it turned out, Microsoft was able to beat OS/2 by withholding Windows 95 licenses for IBM PCs, as was documented in Judge Thomas Penfield Jackson's findings of fact. With OS/2 out of the way, Microsoft could put OFS on the back burner in order to address other pressing threats to its desktop monopoly.
Now OFS is back. Allchin stated last month that OFS is scheduled to go into the next major release of Windows, code-named Longhorn. (Given the lesson of the Monty Python sketch, no doubt Microsoft is referring to the mild cheddar cheese called Longhorn, and not to cattle.) Call me a paranoid cynic, but I'm betting Microsoft is resurrecting OFS as a means to make data more accessible to users while making it less accessible to developers. This would stifle competition from one or more products or technologies, most likely competing productivity applications.
Having said that, an object store based on SQL Server technology is a good idea, though not necessarily the best design for a file system. I've advocated this approach for a long time, and I don't intend to stop just because Microsoft might abuse it. But if you are looking for a much better idea, check out the ReiserFS file system. The file system that Hans Reiser proposes is similar to OFS in one respect: Both are vaporware. But Reiser's vaporware is better than any I've seen in this category.
If I had to summarize Reiser's objectives, I would say that one is to eliminate as many distinctions as possible among various types of files, directories, devices or anything else that can be represented within a file system. The upcoming ReiserFS 4 proposes to do this by turning everything - streams, directories, attributes (time stamps, security settings and others) - into files.
Another goal is to be able to search the file system without having to impose relational database structure upon it. Reiser's examples don't always support his case, but the weakness is in his examples, not his principles. For example, he conjures a story in which you can't search for Santa Claus because the arbitrary structure of the database makes it nearly impossible to define the relationship among Santa, reindeer and chimneys. What Reiser neglects to mention is that a brute-force search engine covers a multitude of structural sins. Unfortunately, the price you'd pay in performance, complexity and disk space (or some combination of these) would outweigh any benefits you'd reap by imposing an arbitrary structure on the file system and then working around the limitations with brute force.
In the end, Reiser's conclusions are perfectly valid. I hope we see the vapor condense into reality before the cheese shop gets its next shipment.
Nicholas Petreley is a computer consultant and author in Hayward, Calif. He can be reached at nicholas@petreley.com.

Read more about Operating Systems in Computerworld's Operating Systems Topic Center.



Our Commenting Policies