Steve: Developing on the Edge - WinFS: "time to knife the baby"

Steve: Developing on the Edge

Thoughts on development, Web-services, technology and mountains.

25Jun
Sun2006

WinFS: "time to knife the baby"

So, WinFS is dead, and the post mortems are out

This is a good thing. It's good for MS because they can stop pouring money into an ocean-boiler of a project, and reassign the engineers to something useful. It also means that anything that stepped near the turf of WinFS won't get perceived as an attempt to undermine the project and get stamped on (*). Best yet, it implies that windows developers should stop holding back for more future releases of stuff, and focus on what Vista is bringing, as that is going to be (relatively) stable for a few years. Indeed, if they code for the Avalon bits that already run on WinXP, .NET 2.0 developers can write apps to ship today, instead of holding off for Vista.

Its good for end users because you don't really need a new and not yet stable filesystem. NTFS was unstable until win2k shipped; I lost two disk partitions simultaneously once to what was probably a race condition bug on a two CPU system. Remember how we didn't get a disk defragmenter until win2K?

It's good for OSS people because yet another windows filesystem was not what we needed. NTFS support on linux isn't great at present; WinFS would be even worse. And, because Vista now won't have a fancy new filesystem with slick relational bits in it, apps that target classic filesystems in a bid to be portable won't appear drastically out of place on Vista. You won't get an obvious split in functionality between winFS apps and "legacy apps".

Now, if WinFS was ready to ship with Vista, and Vista had shipped a few years back, well, perhaps it would have been a good strategic move for MS. A new filesystem which lets your app integrate better, but at the expense of no portability to any platform, and a built in need for Vista and not older versions of windows, versions which are Microsoft's real competitors. But it was precisely the need to add radical features like a new semantic-webby (*) filesystem that caused Vista to slip so badly. The strategic goal "own the files" got in the way of the real one of "ship early and make money".

At the same time, I have to mourn slightly the death of an old idea. WinFS is descended from CairoFS, which was meant to present the OLE2 filesystem interface directly to apps. Instead of saving to OLE2 Structured Storage files (horrible bloated FAT-in-file files), the filesystem itself would be aware of what was going on. manage links between objects saved inside different documents. Furthermore, it would be nice to have a transacted filesystem. Copy on deploy would work properly for more than one file if you could queue up a list of file operations in a single transaction, then commit them atomically. Look how SVN gains from this, and imagine it in your filesystem. I even think integrated metadata could have interesting uses. If MS cannot roll out a radically different filesystem for the desktop, who can?

(*) Trivia note. We were at Chepstow Castle today. The verb to undermine comes from castle siege techniques, where you built a tunnel under a castle tower and then collapsed the mine (by filling it with flammable materials, lighting it and burning down the supports that were holding the mine up. Castles with square corners were more vulnerable to undermining, apparently; I guess they have different load characteristics

semantic-webby, this should mean the same about "semantic web" as "enterprisey" does for enterprises.

Posted by steve at 19:39

Comments

new comment thread


	WinFS - not just a new file systemreply to this thread On 26 June 2006 at 03: 12 Patrick Logan commented: The problems with WinFS: 1. It was trying to do too much, certainly too much at once. No layering of dependencies. 2. It was ignoring the web too much. Yes, it has some network capabilities. But it was mainly a marginally multi-user database/file system from the early 1990's. MS should have focused on making a real database really easy to use by end users and programmers over the web. http://patricklogan.blogspot.com/2004/02/meta-stuff-and-winfs.html
	Seems that you don't know much about WinFSreply to this thread On 26 June 2006 at 09: 44 Fduch commented: "Indeed, if they code for the Avalon bits that already run on WinXP, .NET 2.0 developers can write apps to ship today, instead of holding off for Vista." -Bullshit. WinFS beta works on XP. So developers can code for the WinFS bits that already run on WinXP, .NET 2.0 developers can write apps to ship today, instead of holding off for Vista. "It's good for OSS people because yet another windows filesystem was not what we needed. " - How did you bring OSS here?? And you know.... WinFS is not entire new incompatible file system. it's metadata API layer relying on EXISTING NTFS features to STORE data. And new API to work with it.
	Transactional NTFSreply to this thread On 26 June 2006 at 17: 28 Neil commented: "Furthermore, it would be nice to have a transacted filesystem." Indeed, it would be nice: Vista will include Transactional NTFS (TxF), which allows atomic transactions over multiple files. See http://en.wikipedia.org/wiki/Transactional_NTFS for more information.
	Square towers and medieval miningreply to this thread On 27 June 2006 at 21: 32 Reinout van Rees commented: The problem with the square towers is that they have a clear edge, so something where you tunnel your way to and start knocking out the stones on the edge. That's something you can't easily do with round towers.