What is the purpose of Ext4?
|
| Author | Content |
|---|---|
| Sachankara Jun 30, 2006 2:14 PM EDT |
Perhaps not the best way to introduce oneself and my views, but here it goes: As I wrote in the topic - what is the purpose of Ext4? Seriously? When your fine 5 PiB Ext4 database server crashes and it starts to run fsck on the next boot, you'll just beg to wonder why your server is down for a week or more before you can use it again. There are other file systems out there which are much more competent for large amounts of data. Why not spend all that developer time on XFS/Reiser4/JFS/etc? (Of course, the developers are free to choose for themselves.) Ext3 is just fine for small amounts of data, but not for huge. Sure, Ext3 is quite stable, that's for sure. Ext4 necessarily won't be for a while. My personal opinion is that if one wants to see Ext3/Ext4 as a flexible server file system, more development time must be spent on fixing thing such as dynamic resizing. What good is a server file system if you cannot resize it safely when the server is running? |
| tuxchick2 Jun 30, 2006 2:25 PM EDT |
What I wonder is ZFS going to be ported to other platforms? All the usual filesystems, XFS/JFS/Reiser/Ext aren't much different from each other. But ZFS is truly radical. To quote from the great tech author Carla Schroder: "You know, all those filesystem utilities you've been relying on all these years? All that fsck, dump, restore, mkfs, tunefs, and their Ext3/JFS/XFS/ReiserFS/UFS counterparts; volume managers like EVM and LVM; raidtools; rysnc; quota; fdisk, and all the rest of the baggage you've been forced to lug around just to coax filesystems into a semblance of usefulness - using ZFS means you can dump them all. ZFS was written from the ground up to meet modern needs." http://www.serverwatch.com/tutorials/article.php/3612066 I might even install OpenSolaris just to see what ZFS is like in action. |
| sbergman27 Jun 30, 2006 2:43 PM EDT |
I believe the purpose of Ext4 is to introduce changes which are too disruptive to even think about applying to ext3. Ext3 does not normally require a complete fsck, of course. But as we all know, full fsck time does come around eventually. "Continuation Inodes" ( http://lkml.org/lkml/2006/6/9/476 ) are exactly the sort of thing that could not go into ext3, but could go into ext4. This, I believe, might address the fsck speed problem you mention. Ext3 filesystems can already be grown online. They cannot be shrunk, however. This is another thing that could be more easily addressed in ext4 than in ext3. |
| sbergman27 Jun 30, 2006 2:58 PM EDT |
Carla, There is no way in hell that ZFS proper would make it into the Linux kernel. As Andrew put it, it is a rampant layering violation. But ZFS features? That's another story, altogether. Of course, to get all those features would take a lot more than an ext3/ext4 fork. VFS would need some fairly radical changes, as well. Which is fine. If the stable kernel version is locked at 2.6, the subsystems need to be free to fork for purposes of radical development, lest development stagnate for fear of making changes that are too disruptive to the installed base. We'll just have to ride out the inevitable news reports that the Linux kernel itself has "Fragmented Like UNIX Did (tm)" . |
| tuxchick2 Jun 30, 2006 3:12 PM EDT |
" There is no way in hell that ZFS proper would make it into the Linux kernel. As Andrew put it, it is a rampant layering violation." Can you elaborate on that, or point to some references? From what I've seen, it's pretty darned impressive. Anything that makes life easier for sysadmins gets my vote. |
| jimf Jun 30, 2006 3:24 PM EDT |
Well, a couple of things here. What is applicable and desirable for the desktop, is not necessarily applicable or desirable for the server. Right now reiserFS is probably the best choice for a desktop with a good balance of speed and security. That isn't likely to change soon, and since IBM/SuSe has taken over the stewardship, reiser3 should be around a long time. All my buddys running heavy duty server stuff are using ext3 and wouldn't touch anything else. For them it's the proven and reliable solution which will handle relatively large volumes... But as they say in the article, that is where ext4 comes in. Server apps may very soon 'need' the ability to access very large volumes and deal with the problems that presents. My server Admin friends shudder at the mention of Reiser4, Apparently, Hans Reiser has a horrible track record in support of any of his creations. Some still won't use reiser3 because of its' problems in early versions. Still, DARPA is now sponsoring Reiser4 as is Linspire, so maybe that does have a chance. ZFS sounds interesting, but is 1) proprietary, and, 2) what would it take to work with the Linux kernel... ? Personally, I think this one is a political hot potato. Any solution is going to have to have wide support, and it looks to me like a redesigned ext3 or more likely ext4 will become the mainline solution for server use. |
| sbergman27 Jun 30, 2006 3:29 PM EDT |
> Can you elaborate on that, or point to some references? Of course: http://lkml.org/lkml/2006/6/9/389 I agree completely on the ease of use thing. But it seems to me that the best solution is to fork the appropriate kernel subsystems to add the needed features, and then improve the user space tools to integrate said features into a useful utility. Believe me, I've been frustrated enough by the disparity between LVM tools, raid tools, fs tools, etc. But kitchen-sinking it all into the filesystem doesn't seem like the best long term solution. |
| Sachankara Jun 30, 2006 9:17 PM EDT |
"Ext3 file systems can already be grown online. They cannot be shrunk, however. This is another thing that could be more easily addressed in ext4 than in ext3." Well, you can do online resizing with a fairly unmaintained kernel patch. It needs to be in the main line kernel if you want to be sure that the feature is available for you at "all time". XFS can do online growing of file systems right out of the box - same with ReiserFS and most likely JFS (but I haven't tried). Sure, it could very well be implemented in Ext4, but they're still lagging behind other file systems. Though it would be a welcomed feature none-the-less. |
Posting in this forum is limited to members of the group: [ForumMods, SITEADMINS, MEMBERS.]
Becoming a member of LXer is easy and free. Join Us!