Microsoft Doesn’t Understand (part 42)
A few days ago Frans Bouma asked if the (new?) WinFS delays were
politics or incompetency, pointing to ReiserFS version 4 as an example of a fast, meta-data lovin’ next generation file system. ReiserFS version 4 has been developed and released while WinFS is keeps on getting delayed. Anyway, Robert Scoble linked to this and asked
Think about the engineering problems there. Does Photoshop run on Reiser? Yet it’ll need to run on WinFS cause the market expects that of Windows apps.
Huh?
A while back (several years) Microsoft made a big splash by announcing they were going to have a booth at one of the Linux Shows (don’t remember if it was a Linux
World or what) and everyone was amazed that such a bold move was to be made. The impression that Microsoft had was that “Linux people” just didn’t understand that Microsoft wasn’t a big bad guy, but was there to help. Roberts comments above about Photoshop and ReiserFS indicate that the opposite is true.
- Photoshop runs on linux thanks to Code Weavers Crossover Office.
- The model under linux is that Filesystems and physical devices are pretty independant of the user experience. IE: it doesn’t matter if I run ext3 or ReiserFS or XFS, or if I’m running on one physical disk or 10, it’s just a filesystem, and they’re just files.
Of course, one thing that Robert brought up in my mind (unintentionally I’m sure) is how the different development models work. In the Linux and Open Source world it is possible for a completely separate company to develop something so “close to the metal” if they have the experience because all the APIs, source code, and tools are available free.
If a company wanted to develop a WinFS-type filesystem for Windows would they be able to? Could they give it away for free? I doubt it. License fees, cost of getting Shared Source, etc etc would make it pretty much impossible. Course I might be misinformed, someone prove me wrong.
If I recall correctly – and I may not – there was an ORA book, “NTFS Internals” or some such (one of the blue volumes, not the pink NT ones) that had the details on how to implement a proc-like FS for WinNT.
free but not Free is probably OK for kernel-level stuff.
I’ve also been informed that the IFS dev kit is available, my mistake. However, you also don’t see user created filesystems out there (though I understand they do exist). My thoughts that it’s due to being afraid of being sued or licensing costs, or just the whole “we love microsoft, they’ll do it for us” attitude.
Well, the majority of people use NTFS the same reason why they use IE, they have no choice. Actually they have the choice between NTFS and VFAT if they do their own WinXP (assuming we’re talking desktop users here) install but of course we know how many end users do that 🙂 So really you use the filesystem that is on the computer that’s set up by dell/compaq/whoever.
Do you need different filesystems on linux? Well, each has their own strengths and weaknesses. Just like under windows people aren’t going out and writing the filesystems willy nilly like they are say, window managers. The majority of filesystems are ported over because they are industrial strength (ie: XFS and JFS), are an upgrade to an old version (ext3 is a journalling version of ext2), or are “new and innovative” (reiserfs and it’s plugin system). NTFS is fine as a filesystem, I’m not slagging it. It’s the default that most people will PROBABLY use. Just like ext3 or reiser are the default that the desktop linux user will PROBABLY use (installing from redhat/mandrake/etc).
You should also re-read my post… I wasn’t slagging windows for it’s problems technically (for a change :), but slagging robert for his complete lack of knowledge and FUD spreading, and questioning the politics behind the lack of choice in filesystems that windows has if the ability to write your own and say, give people WinFS before MS can deliver. The answer seems to be that windows users don’t like choice, and will be happy with what they’re given for the most part.
Or perhaps because NTFS, in itself, is a great journalling file system?
You can add the functionality you want as a file system filter on top of NTFS. Don’t like that? Write your own using the IFS kit. Want Reiser? Get off your ass and port it.
This is an argument which I think is futile. You can’t compare ReiserFS to WinFS in a vacuum. It’s not the same thing. The feature sets are different. The corporate politics are different. The amount of regression testing is different (remember reiser4 is NOT as stable as 3 on the latest 2.6x kernels). Hell, the sheer size and complexity of the operating system and the required API hooks are different.
If you WANT to write a FS for Windows you can. And many do. There are lots of examples of specialized encrypted file systems as an example that use the IFS kit to meet the vendor’s own needs. And funky redirectors. And reparsers and filters.
Here is a question though. Notice that the MAJORITY of people use NTFS on Windows. Because it WORKS. And works WELL. And has been for over 15 years. Why they hell is there so many different ones on Linux? Why do we NEED ext3 and reiser AND XFS AND… shall I go on?
One of Linux’s greatest strengths is also its biggest weakness. So much segmentation and multiple development efforts erode the amount of development time and testing that could go into a single file system that could meet everyone’s needs. Open source is about freedom, I grant you that. But it doesn’t mean you SHOULD break off and write your own FS ‘just cuz’ today you don’t like FS-versionX. I look at what Microsoft is doing with Transactional NTFS and realize… ‘they get it’. They realize NTFS rocks, but it could be better with transactional support. And where NTFS has in the past been weak (like 64K sequential writes… a limitation that is over 15 years old now) they have removed that to allow for up to 4 GIG writes.
Slag windows for where it actually faults. Not in areas it doesn’t. Party on if you feel like porting reiser to Windows. I’ll stick with NTFS.