Design documents and specification of Ext4 ?

[quanganht]

Do u think Btrfs is better than ext4 in term of speed and stability? The two file system seems to have the same amount of documents anyway. So I'm looking for faster design, the faster is the better.

[Love4Boobies]

The two file systems struggle to achieve different goals. ext4 is probably more suitable for home systems while Btrfs for servers and perhaps even more distributed environments. If it's speed rather than features you're looking for, they are both in the wrong direction.

[NickJohnson]

For a first (or probably up to third) filesystem, I doubt ext4 is the best way to go - ext2 is much simpler. Afaik, the Linux folks didn't even have a stable implementation of ext4 until a few months ago. If you're not GPLed, and you want speed through complexity anyway, you might as well go with ZFS, which iirc is much better documented.

[quanganht]

Is ZFS the fastest one?

[Combuster]

Google?


Still, if you want an established, used, fast (not fastest), and documented FS, you might want to check out XFS. But starting simple might be, especially considering your reputation in this thread, a much better idea.

[Love4Boobies]

Is ZFS the fastest one?

Haha, did you google for the slowest ones around or are you trying all the ones you know on us?

[quanganht]

I will not try something like ext or FAT16. Actually I (we) have a long term plan so I think it's better to choose a good FS and stick with it until the OS become mature.

[Love4Boobies]

Then try ext2 and/or FAT32. I don't even know any OS that implements ext support.

[jvff]

Hi,

I would suggest ext2. It's not the fastest, but it's relatively fast and it's an easier start for ext3/4. Plus, there are lot's of documentation about ext2. As for ext3, I've only found one useful explanation of the journal implementation, but the Linux code is readable, as long as you know the basics of how file systems are implemented in Linux. I haven't read anything about ext4 yet so I can't help you there... Some books were helpful for me: Unix Filesystems - Evolution, Design and Implementation; and also the VFS, Cache and ext2 chapters of Understanding the Linux Kernel.

If you want the fastest file system, you must first consider where it's going to run. Either way, there is always a tradeoff. Consider Log file-systems, they provide probably the fastest write speeds for HDs (no seeking), however for reading it might be an entirely different story. If you want fast read speeds, tree organization certainly helps a lot (like Btrfs), however write speeds might suffer a little. If you truly want the fastest, you'll probably have to waste some space anyway...

Hope this helps, and good luck!

JVFF