1. Compatibility
Any existing Ext3 filesystem can be migrated to Ext4 with an easy procedure which consists in running a couple of commands in read-only mode (described in the next section). This means that you can improve the performance, storage limits and features of your current filesystems without reformatting and/or reinstalling your OS and software environment. If you need the advantages of Ext4 on a production system, you can upgrade the filesystem. The procedure is safe and doesn't risk your data (obviously, backup of critical data is recommended, even if you aren't updating your filesystem :). Ext4 will use the new data structures only on new data, the old structures will remain untouched and it will be possible to read/modify them when needed. This means, that, of course, that once you convert your filesystem to Ext4 you won't be able to go back to Ext3 again (although there's a possibility, described in the next section, of mounting a Ext3 filesystem with Ext4 without using the new disk format and you'll be able to mount it with Ext3 again, but you lose many of the advantages of Ext4).
2. Bigger filesystem/file sizes
Currently, Ext3 support 16 TB of maximum filesystem size, and 2 TB of maximum file size. Ext4 adds 48-bit block addressing, so it will have 1 EB of maximum filesystem size and 16 TB of maximum file size. 1 EB = 1,048,576 TB (1 EB = 1024 PB, 1 PB = 1024 TB, 1 TB = 1024 GB). Why 48-bit and not 64-bit? There are some limitations that would need to be fixed before making Ext4 fully 64-bit capable, which have not been addressed in Ext4. The Ext4 data structures have been designed keeping this in mind, so a future update to Ext4 will implement full 64-bit support at some point. 1 EB will be enough (really :)) until that happens. (Note: The code to create filesystems bigger than 16 TB is -at the time of writing this article- not in any stable release of e2fsprogs. It will be in future releases.)
3. Sub directory scalability
Right now the maximum possible number of sub directories contained in a single directory in Ext3 is 32000. Ext4 breaks that limit and allows a unlimited number of sub directories.
4. Extents
The traditionally Unix-derived filesystems like Ext3 use a indirect block mapping scheme to keep track of each block used for the blocks corresponding to the data of a file. This is inefficient for large files, specially on large file delete and truncate operations, because the mapping keeps a entry for every single block, and big files have many blocks -> huge mappings, slow to handle. Modern filesystems use a different approach called "extents". An extent is basically a bunch of contiguous physical blocks. It basically says "The data is in the next n blocks". For example, a 100 MB file can be allocated into a single extent of that size, instead of needing to create the indirect mapping for 25600 blocks (4 KB per block). Huge files are split in several extents. Extents improve the performance and also help to reduce the fragmentation, since an extent encourages continuous layouts on the disk.
5. Multiblock allocation
When Ext3 needs to write new data to the disk, there's a block allocator that decides which free blocks will be used to write the data. But the Ext3 block allocator only allocates one block (4KB) at a time. That means that if the system needs to write the 100 MB data mentioned in the previous point, it will need to call the block allocator 25600 times (and it was just 100 MB!). Not only this is inefficient, it doesn't allow the block allocator to optimize the allocation policy because it doesn't knows how many total data is being allocated, it only knows about a single block. Ext4 uses a "multiblock allocator" (mballoc) which allocates many blocks in a single call, instead of a single block per call, avoiding a lot of overhead. This improves the performance, and it's particularly useful with delayed allocation and extents. This feature doesn't affect the disk format. Also, note that the Ext4 block/inode allocator has other improvements, described in detail in this paper.
6. Delayed allocation
Delayed allocation is a performance feature (it doesn't change the disk format) found in a few modern filesystems such as XFS, ZFS, btrfs or Reiser 4, and it consists in delaying the allocation of blocks as much as possible, contrary to what traditionally filesystems (such as Ext3, reiser3, etc) do: allocate the blocks as soon as possible. For example, if a process write()s, the filesystem code will allocate immediately the blocks where the data will be placed - even if the data is not being written right now to the disk and it's going to be kept in the cache for some time. This approach has disadvantages. For example when a process is writing continually to a file that grows, successive write()s allocate blocks for the data, but they don't know if the file will keep growing. Delayed allocation, on the other hand, does not allocate the blocks immediately when the process write()s, rather, it delays the allocation of the blocks while the file is kept in cache, until it is really going to be written to the disk. This gives the block allocator the opportunity to optimize the allocation in situations where the old system couldn't. Delayed allocation plays very nicely with the two previous features mentioned, extents and multiblock allocation, because in many workloads when the file is written finally to the disk it will be allocated in extents whose block allocation is done with the mballoc allocator. The performance is much better, and the fragmentation is much improved in some workloads.
7. Fast fsck
Fsck is a very slow operation, especially the first step: checking all the inodes in the file system. In Ext4, at the end of each group's inode table will be stored a list of unused inodes (with a checksum, for safety), so fsck will not check those inodes. The result is that total fsck time improves from 2 to 20 times, depending on the number of used inodes (http://kerneltrap.org/Linux/Improving_fsck_Speeds_in_Ext4). It must be noticed that it's fsck, and not Ext4, who will build the list of unused inodes. This means that you must run fsck to get the list of unused inodes built, and only the next fsck run will be faster (you need to pass a fsck in order to convert a Ext3 filesystem to Ext4 anyway). There's also a feature that takes part in this fsck speed up - "flexible block groups" - that also speeds up filesystem operations.
8. Journal checksumming
The journal is the most used part of the disk, making the blocks that form part of it more prone to hardware failure. And recovering from a corrupted journal can lead to massive corruption. Ext4 checksums the journal data to know if the journal blocks are failing or corrupted. But journal checksumming has a bonus: it allows one to convert the two-phase commit system of Ext3's journaling to a single phase, speeding the filesystem operation up to 20% in some cases - so reliability and performance are improved at the same time. (Note: the part of the feature that improves the performance, the asynchronous logging, is turned off by default for now, and will be enabled in future releases, when its reliability improves)
9. "No Journaling" mode
Journaling ensures the integrity of the filesystem by keeping a log of the ongoing disk changes. However, it is know to have a small overhead. Some people with special requirements and workloads can run without a journal and its integrity advantages. In Ext4 the journaling feature can be disabled, which provides a small performance improvement.
10. Online defragmentation
(This feature is being developed and will be included in future releases). While delayed allocation, extents and multiblock allocation help to reduce the fragmentation, with usage filesystems can still fragment. For example: You write three files in a directory and continually on the disk. Some day you need to update the file of the middle, but the updated file has grown a bit, so there's not enough room for it. You have no option but fragment the excess of data to another place of the disk, which will cause a seek, or allocate the updated file continually in another place, far from the other two files, resulting in seeks if an application needs to read all the files on a directory (say, a file manager doing thumbnails on a directory full of images). Besides, the filesystem can only care about certain types of fragmentation, it can't know, for example, that it must keep all the boot-related files contiguous, because it doesn't know which files are boot-related. To solve this issue, Ext4 will support online fragmentation, and there's a e4defrag tool which can defragment individual files or the whole filesystem.
11. Inode-related features
Larger inodes, nanosecond timestamps, fast extended attributes, inodes reservation...
* Larger inodes: Ext3 supports configurable inode sizes (via the -I mkfs parameter), but the default inode size is 128 bytes. Ext4 will default to 256 bytes. This is needed to accommodate some extra fields (like nanosecond timestamps or inode versioning), and the remaining space of the inode will be used to store extend attributes that are small enough to fit it that space. This will make the access to those attributes much faster, and improves the performance of applications that use extend attributes by a factor of 3-7 times.
* Inode reservation consists in reserving several inodes when a directory is created, expecting that they will be used in the future. This improves the performance, because when new files are created in that directory they'll be able to use the reserved inodes. File creation and deletion is hence more efficient.
* Nanoseconds timestamps means that inode fields like "modified time" will be able to use nanosecond resolution instead of the second resolution of Ext3.
12. Persistent preallocation
This feature, available in Ext3 in the latest kernel versions, and emulated by glibc in the filesystems that don't support it, allows applications to preallocate disk space: Applications tell the filesystem to preallocate the space, and the filesystem preallocates the necessary blocks and data structures, but there's no data on it until the application really needs to write the data in the future. This is what P2P applications do in their own when they "preallocate" the necessary space for a download that will last hours or days, but implemented much more efficiently by the filesystem and with a generic API. This have several uses: first, to avoid applications (like P2P apps) doing it themselves inefficiently by filling a file with zeros. Second, to improve fragmentation, since the blocks will be allocated at one time, as contiguously as possible. Third, to ensure that applications has always the space they know they will need, which is important for RT-ish applications, since without preallocation the filesystem could get full in the middle of an important operation. The feature is available via the libc posix_fallocate() interface.
13. Barriers on by default
This is an option that improves the integrity of the filesystem at the cost of some performance (you can disable it with "mount -o barrier=0", recommended trying it if you're benchmarking). From this LWN article: "The filesystem code must, before writing the [journaling] commit record, be absolutely sure that all of the transaction's information has made it to the journal. Just doing the writes in the proper order is insufficient; contemporary drives maintain large internal caches and will reorder operations for better performance. So the filesystem must explicitly instruct the disk to get all of the journal data onto the media before writing the commit record; if the commit record gets written first, the journal may be corrupted. The kernel's block I/O subsystem makes this capability available through the use of barriers; in essence, a barrier forbids the writing of any blocks after the barrier until all blocks written before the barrier are committed to the media. By using barriers, filesystems can make sure that their on-disk structures remain consistent at all times."
