Tuxera Claims NTFS Is The Fastest File-System For Linux

[allquixotic]

Linux is also used as a base for firmware in a lot of consumer electronics, e.g., nowadays many modern television sets have ntfs support, so users can share movies between their PC and TV through a USB drive.

Since Microsoft Windows is still the dominating desktop OS and it supports a rather limited set of file systems of which FAT's 4GB file size limit can be problematic for content in HD, the choice to also add ntfs support is easy.

This is a very good point. Supporting USB/FireWire mass storage devices mounted as NTFS is an important component of many consumer electronics. Maybe Tuxera's target market is those manufacturers. This would also agree with the reason why Anton's example data was on an embedded (probably ARM) system; he's probably sitting there with a few embedded reference systems, making code changes and testing the performance delta on those systems. In light of your insight, I doubt even more that he cares at all how these tweaks affect data integrity or multi-core x86 desktops and servers.

[movieman]

If you don't completely commit the journal, the metadata and the file data to the media when userspace calls fsync(), you aren't doing your job as a filesystem.

I disagree: I'd love to have a filesystem that maintains consistency but ignores fsync() on some of my systems. In many cases I care more about the number of writes and/or power usage than whether the data actually got to the disk... flushing it every few minutes would be fine, because I can live with losing a few minutes of data.

The fsync() call is truly evil because it allows individual applications to make decisions that should be made on a system-wide basis. If I'm using a laptop on battery I really don't want the hard drive spinning all the time because some stupid program has decided it wants to call fsync() every few seconds. Similarly, I don't want to burn out my SSD because some stupid program is calling fsync() every few seconds and forcing multiple small writes which could have been accumulated into larger writes.

[FunkyRider]

Sorry but, no matter how good you claim your NTFS driver to be, running a binary blob file system driver on a OSS system DOES NOT MAKE SENSE. I'd rather use the open EXT4 for now and wait for the improvement to be laid out, than strangling myself onto the closed blob driver tree


For GPU driver it is easy, if you announce that you are stopping the support today, I will go buy someone else's GPU the next minute.

For FS driver, when you announce you are stopping support or charge a million, I will have to redo all my systems AND backups, not such a good idea.

[FunkyRider]

Forgot to say to this author:

If you want us to use the NTFS driver then:

GO FORK YOURSELF. (TO BSD/LGPL)

LOL

[allquixotic]

I disagree: I'd love to have a filesystem that maintains consistency but ignores fsync() on some of my systems. In many cases I care more about the number of writes and/or power usage than whether the data actually got to the disk... flushing it every few minutes would be fine, because I can live with losing a few minutes of data.

The fsync() call is truly evil because it allows individual applications to make decisions that should be made on a system-wide basis. If I'm using a laptop on battery I really don't want the hard drive spinning all the time because some stupid program has decided it wants to call fsync() every few seconds. Similarly, I don't want to burn out my SSD because some stupid program is calling fsync() every few seconds and forcing multiple small writes which could have been accumulated into larger writes.

It's not up to you or me or the kernel developers to decide the semantics of fsync(). It's a standard system call supported across many different UNIX distributions, and its semantics are expected to be consistent. The issue isn't debatable.

Even the Linux kernel developers insist that userspace APIs remain rock solid across releases. Except for some of the DRM stuff (which to be fair is only used by libdrm in userspace), the exposed userland APIs of the kernel -- especially those that are standard across many distributions and have years of history -- have remained rock-solid and unchanging throughout the years. It would be particularly surprising if one filesystem doesn't respect fsync() while another does. If you want to break semantics for a call that's been around for 13+ years, you should do it across the board on all filesystems and loudly document it on the big horn.

It's all about standards and keeping our promises to userland. Otherwise we may as well arbitrarily add parameters to fsync() with each minor release of the kernel, and expect every user to adjust.

You're welcome to use libnofsync if you like to lose data, but the default shouldn't be to go against long-standing standards.

[XorEaxEax]

NTFS-3G is already quite fast; my Freecom 80GB USB disk works with 25 MB/s write speed using Linux. On Windows it writes at 20 MB/s, using exactly the same laptop. (The limit is not the disk on the other side of the copy; the Western Digital 250GB of my laptop can read files much faster)

Well, I've never considered ntfs-3g fast, but then again it runs in userspace so that obviously makes it slower than it has to be. Sadly though it's mimics original NTFS too well in my opinion, given that it fragments like a bitch just like the original. Someone in this thread mentioned that it doesn't, lol wut? Have they actually written a number of large files and then examined the fragmentation? I did tests on a fully defragmented partition (used jkdefrag with 'move files together' algorithm) with 120gb free (in a large chunk) where I wrote 5 files ranging from 10gb to 4gb and ntfs_3g fragmented them like crazy. Sure it stays true to the original in this way but I really would have wished that they'd (ntfs_3g) improve on this fallacy of NTFS.

[siride]

Very simple - go to xxx site and see how your windows's falling down.

I have and it doesn't.

Listen up fanboys: it's not 1998 anymore. Windows is reasonably secure, even for regular users (I don't get calls from my mom asking me to fix a virus anymore since she's been on Vista with MS Security Essentials). Windows has a modern kernel and a modern filesystem. All of these tired arguments from the Windows 3.1 and 95 era need to die. They needed to die years ago.

[smitty3268]

any unix based system can run desktop games securely because it works exactly the same way as every other program that ever runs - the game data is stored in a place where only root has access to read and write. it doesn't matter if you have a gui or not.

You apparently don't realize Windows does the exact same thing. It has since Vista came out. This isn't Windows 98 anymore where everyone runs as admin. I don't think I've ever seen a user running Win7 as admin, and it was only somewhat common on Vista while all the programs requiring admin access got rewritten to not require it.

who the hell uses windows without a gui and has time (or the authority/access) to compile everything they want in a cli? that literally isn't useless but comparatively, it's very very useless and pointless.

I assumed we were talking about enterprise servers here, because the topic was how awesome Linux was that it didn't require you to run X. Now it sounds like you're talking about desktop use, which certainly requires X just as much as Windows requires a GUI in such a situation. I'm not sure i see the difference here, other than Linux allowing a bit more flexibility while Windows is a more standard shrink-wrapped version.

[smitty3268]

Does Windows have a CLI option to replace their GUI configuration dialogs?
Like the whole Control Panel is all GUI based, is there some way to do that configuration from a CLI? It might be possible to do in the registry, but the whole registry is a screwed up mess. Is there even a CLI registry browser in Windows. Regedit.exe is a GUI program. :/
I do know that you can edit the registry through Batch, but you have to know ahead of time what keys you're editing. I'm not aware of a browser like Regedit where you can just peruse the registry.

Another problem with running Windows without a GUI is that it'd be a pain in the ass to install software, not just because most programs require a GUI, but because there isn't a package manager and there isn't a built-in wget/curl utility for downloading executables. So you would use another computer and a browser to download the file then transfer it to your server.

The standard CMD shell just seems like it's lacking too much to be the sole interface to a system.

This is what powershell was designed for. It's like Python integrated into the OS.

[smitty3268]

I'm a linux fanboy

and i wish it were still true that Linux had a big security advantage. It's not. Those saying it does are like the Mac fanboys claiming that the lack of viruses prove how superior OSX security is, which is a joke.

Windows and Linux security is on par with each other, and pretending otherwise by burying our heads in the sand just means that it won't ever get any better.

By far the biggest difference is that Linux tends to be run by people who understand computers, or by people who had it setup by people who understand computers. That means that it tends to be configured in relatively sane ways and used by people who are smart enough not to enter their root password to view an email attachment just because it says it has nude pictures in it. Windows is used by idiots worldwide who do exactly that, and there's nothing Linux can do to protect users like that from themselves anymore than Windows can.